From b5b4061887617eac28c2421c955aeb9bbaa9f79b Mon Sep 17 00:00:00 2001 From: nullquant Date: Fri, 3 May 2024 12:13:56 +0300 Subject: [PATCH] PowerPaint preparation --- brushnet/brushnet_ca.py | 164 +- brushnet/pipeline_brushnet.py | 1284 -------------- brushnet/pipeline_brushnet_powerpaint.py | 1316 -------------- brushnet/pipeline_brushnet_sd_xl.py | 1539 ----------------- ....py => pipeline_powerpaint_brushnet_ca.py} | 116 +- brushnet/powerpaint.json | 57 + brushnet/powerpaint_utils.py | 496 ++++++ brushnet/utils.py | 528 ++++++ brushnet_nodes.py | 44 +- 9 files changed, 1266 insertions(+), 4278 deletions(-) delete mode 100644 brushnet/pipeline_brushnet.py delete mode 100644 brushnet/pipeline_brushnet_powerpaint.py delete mode 100644 brushnet/pipeline_brushnet_sd_xl.py rename brushnet/{pipeline_brushnet_powerpaint_ca.py => pipeline_powerpaint_brushnet_ca.py} (96%) create mode 100644 brushnet/powerpaint.json create mode 100644 brushnet/powerpaint_utils.py create mode 100644 brushnet/utils.py diff --git a/brushnet/brushnet_ca.py b/brushnet/brushnet_ca.py index 558f249..10366de 100644 --- a/brushnet/brushnet_ca.py +++ b/brushnet/brushnet_ca.py @@ -3,20 +3,20 @@ from typing import Any, Dict, List, Optional, Tuple, Union import torch from torch import nn -from torch.nn import functional as F -from ..configuration_utils import ConfigMixin, register_to_config -from ..utils import BaseOutput, logging -from .attention_processor import ( +from diffusers.configuration_utils import ConfigMixin, register_to_config +from diffusers.utils import BaseOutput, logging +from diffusers.models.attention_processor import ( ADDED_KV_ATTENTION_PROCESSORS, CROSS_ATTENTION_PROCESSORS, AttentionProcessor, AttnAddedKVProcessor, AttnProcessor, ) -from .embeddings import TextImageProjection, TextImageTimeEmbedding, TextTimeEmbedding, TimestepEmbedding, Timesteps -from .modeling_utils import ModelMixin -from .unets.unet_2d_blocks import ( +from diffusers.models.embeddings import TextImageProjection, TextImageTimeEmbedding, TextTimeEmbedding, TimestepEmbedding, Timesteps +from diffusers.models.modeling_utils import ModelMixin + +from .unet_2d_blocks import ( CrossAttnDownBlock2D, DownBlock2D, UNetMidBlock2D, @@ -27,7 +27,7 @@ from .unets.unet_2d_blocks import ( MidBlock2D ) -from .unets.unet_2d_condition import UNet2DConditionModel +from .unet_2d_condition import UNet2DConditionModel logger = logging.get_logger(__name__) # pylint: disable=invalid-name @@ -36,7 +36,7 @@ logger = logging.get_logger(__name__) # pylint: disable=invalid-name @dataclass class BrushNetOutput(BaseOutput): """ - The output of [`BrushNetCAModel`]. + The output of [`BrushNetModel`]. Args: up_block_res_samples (`tuple[torch.Tensor]`): @@ -58,9 +58,9 @@ class BrushNetOutput(BaseOutput): mid_block_res_sample: torch.Tensor -class BrushNetCAModel(ModelMixin, ConfigMixin): +class BrushNetModel(ModelMixin, ConfigMixin): """ - A BrushNetCA model. + A BrushNet model. Args: in_channels (`int`, defaults to 4): @@ -142,22 +142,17 @@ class BrushNetCAModel(ModelMixin, ConfigMixin): flip_sin_to_cos: bool = True, freq_shift: int = 0, down_block_types: Tuple[str, ...] = ( - # "CrossAttnDownBlock2D", - # "DownBlock2D", - # "DownBlock2D", - # "DownBlock2D", "CrossAttnDownBlock2D", "CrossAttnDownBlock2D", "CrossAttnDownBlock2D", "DownBlock2D", ), - mid_block_type: Optional[str] = "UNetMidBlock2DCrossAttn", #"UNetMidBlock2D", + mid_block_type: Optional[str] = "UNetMidBlock2DCrossAttn", up_block_types: Tuple[str, ...] = ( - # "UpBlock2D", - # "UpBlock2D", - # "UpBlock2D", - # "CrossAttnUpBlock2D", - "UpBlock2D", "CrossAttnUpBlock2D", "CrossAttnUpBlock2D", "CrossAttnUpBlock2D" + "UpBlock2D", + "CrossAttnUpBlock2D", + "CrossAttnUpBlock2D", + "CrossAttnUpBlock2D", ), only_cross_attention: Union[bool, Tuple[bool]] = False, block_out_channels: Tuple[int, ...] = (320, 640, 1280, 1280), @@ -224,7 +219,10 @@ class BrushNetCAModel(ModelMixin, ConfigMixin): conv_in_kernel = 3 conv_in_padding = (conv_in_kernel - 1) // 2 self.conv_in_condition = nn.Conv2d( - in_channels+conditioning_channels, block_out_channels[0], kernel_size=conv_in_kernel, padding=conv_in_padding + in_channels + conditioning_channels, + block_out_channels[0], + kernel_size=conv_in_kernel, + padding=conv_in_padding, ) # time @@ -376,19 +374,19 @@ class BrushNetCAModel(ModelMixin, ConfigMixin): self.brushnet_mid_block = brushnet_block self.mid_block = get_mid_block( - mid_block_type, - transformer_layers_per_block=transformer_layers_per_block[-1], - in_channels=mid_block_channel, - temb_channels=time_embed_dim, - resnet_eps=norm_eps, - resnet_act_fn=act_fn, - output_scale_factor=mid_block_scale_factor, - resnet_time_scale_shift=resnet_time_scale_shift, - cross_attention_dim=cross_attention_dim, - num_attention_heads=num_attention_heads[-1], - resnet_groups=norm_num_groups, - use_linear_projection=use_linear_projection, - upcast_attention=upcast_attention, + mid_block_type, + transformer_layers_per_block=transformer_layers_per_block[-1], + in_channels=mid_block_channel, + temb_channels=time_embed_dim, + resnet_eps=norm_eps, + resnet_act_fn=act_fn, + output_scale_factor=mid_block_scale_factor, + resnet_time_scale_shift=resnet_time_scale_shift, + cross_attention_dim=cross_attention_dim, + num_attention_heads=num_attention_heads[-1], + resnet_groups=norm_num_groups, + use_linear_projection=use_linear_projection, + upcast_attention=upcast_attention, ) # count how many layers upsample the images @@ -397,11 +395,11 @@ class BrushNetCAModel(ModelMixin, ConfigMixin): # up reversed_block_out_channels = list(reversed(block_out_channels)) reversed_num_attention_heads = list(reversed(num_attention_heads)) - reversed_transformer_layers_per_block = (list(reversed(transformer_layers_per_block))) + reversed_transformer_layers_per_block = list(reversed(transformer_layers_per_block)) only_cross_attention = list(reversed(only_cross_attention)) output_channel = reversed_block_out_channels[0] - + self.up_blocks = nn.ModuleList([]) self.brushnet_up_blocks = nn.ModuleList([]) @@ -421,7 +419,7 @@ class BrushNetCAModel(ModelMixin, ConfigMixin): up_block = get_up_block( up_block_type, - num_layers=layers_per_block+1, + num_layers=layers_per_block + 1, transformer_layers_per_block=reversed_transformer_layers_per_block[i], in_channels=input_channel, out_channels=output_channel, @@ -442,8 +440,8 @@ class BrushNetCAModel(ModelMixin, ConfigMixin): ) self.up_blocks.append(up_block) prev_output_channel = output_channel - - for _ in range(layers_per_block+1): + + for _ in range(layers_per_block + 1): brushnet_block = nn.Conv2d(output_channel, output_channel, kernel_size=1) brushnet_block = zero_module(brushnet_block) self.brushnet_up_blocks.append(brushnet_block) @@ -453,7 +451,6 @@ class BrushNetCAModel(ModelMixin, ConfigMixin): brushnet_block = zero_module(brushnet_block) self.brushnet_up_blocks.append(brushnet_block) - @classmethod def from_unet( cls, @@ -464,11 +461,11 @@ class BrushNetCAModel(ModelMixin, ConfigMixin): conditioning_channels: int = 5, ): r""" - Instantiate a [`BrushNetCAModel`] from [`UNet2DConditionModel`]. + Instantiate a [`BrushNetModel`] from [`UNet2DConditionModel`]. Parameters: unet (`UNet2DConditionModel`): - The UNet model weights to copy to the [`BrushNetCAModel`]. All configuration options are also copied + The UNet model weights to copy to the [`BrushNetModel`]. All configuration options are also copied where applicable. """ transformer_layers_per_block = ( @@ -486,14 +483,16 @@ class BrushNetCAModel(ModelMixin, ConfigMixin): conditioning_channels=conditioning_channels, flip_sin_to_cos=unet.config.flip_sin_to_cos, freq_shift=unet.config.freq_shift, - # down_block_types=['CrossAttnDownBlock2D','DownBlock2D','DownBlock2D','DownBlock2D'], - down_block_types=["CrossAttnDownBlock2D", - "CrossAttnDownBlock2D", - "CrossAttnDownBlock2D", - "DownBlock2D",], - mid_block_type='UNetMidBlock2DCrossAttn', - # mid_block_type="UNetMidBlock2DCrossAttn", - # up_block_types=['UpBlock2D','UpBlock2D','UpBlock2D','CrossAttnUpBlock2D'], + # down_block_types=['DownBlock2D','DownBlock2D','DownBlock2D','DownBlock2D'], + down_block_types=[ + "CrossAttnDownBlock2D", + "CrossAttnDownBlock2D", + "CrossAttnDownBlock2D", + "DownBlock2D", + ], + # mid_block_type='MidBlock2D', + mid_block_type="UNetMidBlock2DCrossAttn", + # up_block_types=['UpBlock2D','UpBlock2D','UpBlock2D','UpBlock2D'], up_block_types=["UpBlock2D", "CrossAttnUpBlock2D", "CrossAttnUpBlock2D", "CrossAttnUpBlock2D"], only_cross_attention=unet.config.only_cross_attention, block_out_channels=unet.config.block_out_channels, @@ -522,11 +521,11 @@ class BrushNetCAModel(ModelMixin, ConfigMixin): ) if load_weights_from_unet: - conv_in_condition_weight=torch.zeros_like(brushnet.conv_in_condition.weight) - conv_in_condition_weight[:,:4,...]=unet.conv_in.weight - conv_in_condition_weight[:,4:8,...]=unet.conv_in.weight - brushnet.conv_in_condition.weight=torch.nn.Parameter(conv_in_condition_weight) - brushnet.conv_in_condition.bias=unet.conv_in.bias + conv_in_condition_weight = torch.zeros_like(brushnet.conv_in_condition.weight) + conv_in_condition_weight[:, :4, ...] = unet.conv_in.weight + conv_in_condition_weight[:, 4:8, ...] = unet.conv_in.weight + brushnet.conv_in_condition.weight = torch.nn.Parameter(conv_in_condition_weight) + brushnet.conv_in_condition.bias = unet.conv_in.bias brushnet.time_proj.load_state_dict(unet.time_proj.state_dict()) brushnet.time_embedding.load_state_dict(unet.time_embedding.state_dict()) @@ -534,9 +533,9 @@ class BrushNetCAModel(ModelMixin, ConfigMixin): if brushnet.class_embedding: brushnet.class_embedding.load_state_dict(unet.class_embedding.state_dict()) - brushnet.down_blocks.load_state_dict(unet.down_blocks.state_dict(),strict=False) - brushnet.mid_block.load_state_dict(unet.mid_block.state_dict(),strict=False) - brushnet.up_blocks.load_state_dict(unet.up_blocks.state_dict(),strict=False) + brushnet.down_blocks.load_state_dict(unet.down_blocks.state_dict(), strict=False) + brushnet.mid_block.load_state_dict(unet.mid_block.state_dict(), strict=False) + brushnet.up_blocks.load_state_dict(unet.up_blocks.state_dict(), strict=False) return brushnet.to(unet.dtype) @@ -702,7 +701,7 @@ class BrushNetCAModel(ModelMixin, ConfigMixin): return_dict: bool = True, ) -> Union[BrushNetOutput, Tuple[Tuple[torch.FloatTensor, ...], torch.FloatTensor]]: """ - The [`BrushNetCAModel`] forward method. + The [`BrushNetModel`] forward method. Args: sample (`torch.FloatTensor`): @@ -818,10 +817,9 @@ class BrushNetCAModel(ModelMixin, ConfigMixin): emb = emb + aug_emb if aug_emb is not None else emb # 2. pre-process - brushnet_cond=torch.concat([sample,brushnet_cond],1) + brushnet_cond = torch.concat([sample, brushnet_cond], 1) sample = self.conv_in_condition(brushnet_cond) - # 3. down down_block_res_samples = (sample,) for downsample_block in self.down_blocks: @@ -844,7 +842,6 @@ class BrushNetCAModel(ModelMixin, ConfigMixin): down_block_res_sample = brushnet_down_block(down_block_res_sample) brushnet_down_block_res_samples = brushnet_down_block_res_samples + (down_block_res_sample,) - # 5. mid if self.mid_block is not None: if hasattr(self.mid_block, "has_cross_attention") and self.mid_block.has_cross_attention: @@ -861,7 +858,6 @@ class BrushNetCAModel(ModelMixin, ConfigMixin): # 6. BrushNet mid blocks brushnet_mid_block_res_sample = self.brushnet_mid_block(sample) - # 7. up up_block_res_samples = () for i, upsample_block in enumerate(self.up_blocks): @@ -884,7 +880,7 @@ class BrushNetCAModel(ModelMixin, ConfigMixin): cross_attention_kwargs=cross_attention_kwargs, upsample_size=upsample_size, attention_mask=attention_mask, - return_res_samples=True + return_res_samples=True, ) else: sample, up_res_samples = upsample_block( @@ -892,7 +888,7 @@ class BrushNetCAModel(ModelMixin, ConfigMixin): temb=emb, res_hidden_states_tuple=res_samples, upsample_size=upsample_size, - return_res_samples=True + return_res_samples=True, ) up_block_res_samples += up_res_samples @@ -905,18 +901,36 @@ class BrushNetCAModel(ModelMixin, ConfigMixin): # 6. scaling if guess_mode and not self.config.global_pool_conditions: - scales = torch.logspace(-1, 0, len(brushnet_down_block_res_samples) + 1 + len(brushnet_up_block_res_samples), device=sample.device) # 0.1 to 1.0 + scales = torch.logspace( + -1, + 0, + len(brushnet_down_block_res_samples) + 1 + len(brushnet_up_block_res_samples), + device=sample.device, + ) # 0.1 to 1.0 scales = scales * conditioning_scale - brushnet_down_block_res_samples = [sample * scale for sample, scale in zip(brushnet_down_block_res_samples, scales[:len(brushnet_down_block_res_samples)])] - brushnet_mid_block_res_sample = brushnet_mid_block_res_sample * scales[len(brushnet_down_block_res_samples)] - brushnet_up_block_res_samples = [sample * scale for sample, scale in zip(brushnet_up_block_res_samples, scales[len(brushnet_down_block_res_samples)+1:])] + brushnet_down_block_res_samples = [ + sample * scale + for sample, scale in zip( + brushnet_down_block_res_samples, scales[: len(brushnet_down_block_res_samples)] + ) + ] + brushnet_mid_block_res_sample = ( + brushnet_mid_block_res_sample * scales[len(brushnet_down_block_res_samples)] + ) + brushnet_up_block_res_samples = [ + sample * scale + for sample, scale in zip( + brushnet_up_block_res_samples, scales[len(brushnet_down_block_res_samples) + 1 :] + ) + ] else: - brushnet_down_block_res_samples = [sample * conditioning_scale for sample in brushnet_down_block_res_samples] + brushnet_down_block_res_samples = [ + sample * conditioning_scale for sample in brushnet_down_block_res_samples + ] brushnet_mid_block_res_sample = brushnet_mid_block_res_sample * conditioning_scale brushnet_up_block_res_samples = [sample * conditioning_scale for sample in brushnet_up_block_res_samples] - if self.config.global_pool_conditions: brushnet_down_block_res_samples = [ torch.mean(sample, dim=(2, 3), keepdim=True) for sample in brushnet_down_block_res_samples @@ -930,13 +944,13 @@ class BrushNetCAModel(ModelMixin, ConfigMixin): return (brushnet_down_block_res_samples, brushnet_mid_block_res_sample, brushnet_up_block_res_samples) return BrushNetOutput( - down_block_res_samples=brushnet_down_block_res_samples, + down_block_res_samples=brushnet_down_block_res_samples, mid_block_res_sample=brushnet_mid_block_res_sample, - up_block_res_samples=brushnet_up_block_res_samples + up_block_res_samples=brushnet_up_block_res_samples, ) def zero_module(module): for p in module.parameters(): nn.init.zeros_(p) - return module \ No newline at end of file + return module diff --git a/brushnet/pipeline_brushnet.py b/brushnet/pipeline_brushnet.py deleted file mode 100644 index c2ec676..0000000 --- a/brushnet/pipeline_brushnet.py +++ /dev/null @@ -1,1284 +0,0 @@ -import inspect -from typing import Any, Callable, Dict, List, Optional, Tuple, Union - -import numpy as np -import PIL.Image -import torch -import torch.nn.functional as F -from transformers import CLIPImageProcessor, CLIPTextModel, CLIPTokenizer, CLIPVisionModelWithProjection - -from diffusers.image_processor import PipelineImageInput, VaeImageProcessor -from diffusers.loaders import FromSingleFileMixin, IPAdapterMixin, LoraLoaderMixin, TextualInversionLoaderMixin -from diffusers.models import AutoencoderKL, ImageProjection -from .brushnet import BrushNetModel -from .unet_2d_condition import UNet2DConditionModel -from diffusers.models.lora import adjust_lora_scale_text_encoder -from diffusers.schedulers import KarrasDiffusionSchedulers -from diffusers.utils import ( - USE_PEFT_BACKEND, - deprecate, - logging, - replace_example_docstring, - scale_lora_layers, - unscale_lora_layers, -) -from diffusers.utils.torch_utils import is_compiled_module, is_torch_version, randn_tensor -from diffusers.pipelines.pipeline_utils import DiffusionPipeline, StableDiffusionMixin -from diffusers.pipelines.stable_diffusion.pipeline_output import StableDiffusionPipelineOutput -from diffusers.pipelines.stable_diffusion.safety_checker import StableDiffusionSafetyChecker - - -logger = logging.get_logger(__name__) # pylint: disable=invalid-name - - -EXAMPLE_DOC_STRING = """ - Examples: - ```py - from diffusers import StableDiffusionBrushNetPipeline, BrushNetModel, UniPCMultistepScheduler - from diffusers.utils import load_image - import torch - import cv2 - import numpy as np - from PIL import Image - - base_model_path = "runwayml/stable-diffusion-v1-5" - brushnet_path = "ckpt_path" - - brushnet = BrushNetModel.from_pretrained(brushnet_path, torch_dtype=torch.float16) - pipe = StableDiffusionBrushNetPipeline.from_pretrained( - base_model_path, brushnet=brushnet, torch_dtype=torch.float16, low_cpu_mem_usage=False - ) - - # speed up diffusion process with faster scheduler and memory optimization - pipe.scheduler = UniPCMultistepScheduler.from_config(pipe.scheduler.config) - # remove following line if xformers is not installed or when using Torch 2.0. - # pipe.enable_xformers_memory_efficient_attention() - # memory optimization. - pipe.enable_model_cpu_offload() - - image_path="examples/brushnet/src/test_image.jpg" - mask_path="examples/brushnet/src/test_mask.jpg" - caption="A cake on the table." - - init_image = cv2.imread(image_path) - mask_image = 1.*(cv2.imread(mask_path).sum(-1)>255)[:,:,np.newaxis] - init_image = init_image * (1-mask_image) - - init_image = Image.fromarray(init_image.astype(np.uint8)).convert("RGB") - mask_image = Image.fromarray(mask_image.astype(np.uint8).repeat(3,-1)*255).convert("RGB") - - generator = torch.Generator("cuda").manual_seed(1234) - - image = pipe( - caption, - init_image, - mask_image, - num_inference_steps=50, - generator=generator, - paintingnet_conditioning_scale=1.0 - ).images[0] - image.save("output.png") - ``` -""" - - -# Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.retrieve_timesteps -def retrieve_timesteps( - scheduler, - num_inference_steps: Optional[int] = None, - device: Optional[Union[str, torch.device]] = None, - timesteps: Optional[List[int]] = None, - **kwargs, -): - """ - Calls the scheduler's `set_timesteps` method and retrieves timesteps from the scheduler after the call. Handles - custom timesteps. Any kwargs will be supplied to `scheduler.set_timesteps`. - - Args: - scheduler (`SchedulerMixin`): - The scheduler to get timesteps from. - num_inference_steps (`int`): - The number of diffusion steps used when generating samples with a pre-trained model. If used, - `timesteps` must be `None`. - device (`str` or `torch.device`, *optional*): - The device to which the timesteps should be moved to. If `None`, the timesteps are not moved. - timesteps (`List[int]`, *optional*): - Custom timesteps used to support arbitrary spacing between timesteps. If `None`, then the default - timestep spacing strategy of the scheduler is used. If `timesteps` is passed, `num_inference_steps` - must be `None`. - - Returns: - `Tuple[torch.Tensor, int]`: A tuple where the first element is the timestep schedule from the scheduler and the - second element is the number of inference steps. - """ - if timesteps is not None: - accepts_timesteps = "timesteps" in set(inspect.signature(scheduler.set_timesteps).parameters.keys()) - if not accepts_timesteps: - raise ValueError( - f"The current scheduler class {scheduler.__class__}'s `set_timesteps` does not support custom" - f" timestep schedules. Please check whether you are using the correct scheduler." - ) - scheduler.set_timesteps(timesteps=timesteps, device=device, **kwargs) - timesteps = scheduler.timesteps - num_inference_steps = len(timesteps) - else: - scheduler.set_timesteps(num_inference_steps, device=device, **kwargs) - timesteps = scheduler.timesteps - return timesteps, num_inference_steps - - -class StableDiffusionBrushNetPipeline( - DiffusionPipeline, - StableDiffusionMixin, - TextualInversionLoaderMixin, - LoraLoaderMixin, - IPAdapterMixin, - FromSingleFileMixin, -): - r""" - Pipeline for text-to-image generation using Stable Diffusion with BrushNet guidance. - - This model inherits from [`DiffusionPipeline`]. Check the superclass documentation for the generic methods - implemented for all pipelines (downloading, saving, running on a particular device, etc.). - - The pipeline also inherits the following loading methods: - - [`~loaders.TextualInversionLoaderMixin.load_textual_inversion`] for loading textual inversion embeddings - - [`~loaders.LoraLoaderMixin.load_lora_weights`] for loading LoRA weights - - [`~loaders.LoraLoaderMixin.save_lora_weights`] for saving LoRA weights - - [`~loaders.FromSingleFileMixin.from_single_file`] for loading `.ckpt` files - - [`~loaders.IPAdapterMixin.load_ip_adapter`] for loading IP Adapters - - Args: - vae ([`AutoencoderKL`]): - Variational Auto-Encoder (VAE) model to encode and decode images to and from latent representations. - text_encoder ([`~transformers.CLIPTextModel`]): - Frozen text-encoder ([clip-vit-large-patch14](https://huggingface.co/openai/clip-vit-large-patch14)). - tokenizer ([`~transformers.CLIPTokenizer`]): - A `CLIPTokenizer` to tokenize text. - unet ([`UNet2DConditionModel`]): - A `UNet2DConditionModel` to denoise the encoded image latents. - brushnet ([`BrushNetModel`]`): - Provides additional conditioning to the `unet` during the denoising process. - scheduler ([`SchedulerMixin`]): - A scheduler to be used in combination with `unet` to denoise the encoded image latents. Can be one of - [`DDIMScheduler`], [`LMSDiscreteScheduler`], or [`PNDMScheduler`]. - safety_checker ([`StableDiffusionSafetyChecker`]): - Classification module that estimates whether generated images could be considered offensive or harmful. - Please refer to the [model card](https://huggingface.co/runwayml/stable-diffusion-v1-5) for more details - about a model's potential harms. - feature_extractor ([`~transformers.CLIPImageProcessor`]): - A `CLIPImageProcessor` to extract features from generated images; used as inputs to the `safety_checker`. - """ - - model_cpu_offload_seq = "text_encoder->image_encoder->unet->vae" - _optional_components = ["safety_checker", "feature_extractor", "image_encoder"] - _exclude_from_cpu_offload = ["safety_checker"] - _callback_tensor_inputs = ["latents", "prompt_embeds", "negative_prompt_embeds"] - - def __init__( - self, - vae: AutoencoderKL, - text_encoder: CLIPTextModel, - tokenizer: CLIPTokenizer, - unet: UNet2DConditionModel, - brushnet: BrushNetModel, - scheduler: KarrasDiffusionSchedulers, - safety_checker: StableDiffusionSafetyChecker, - feature_extractor: CLIPImageProcessor, - image_encoder: CLIPVisionModelWithProjection = None, - requires_safety_checker: bool = True, - ): - super().__init__() - - if safety_checker is None and requires_safety_checker: - logger.warning( - f"You have disabled the safety checker for {self.__class__} by passing `safety_checker=None`. Ensure" - " that you abide to the conditions of the Stable Diffusion license and do not expose unfiltered" - " results in services or applications open to the public. Both the diffusers team and Hugging Face" - " strongly recommend to keep the safety filter enabled in all public facing circumstances, disabling" - " it only for use-cases that involve analyzing network behavior or auditing its results. For more" - " information, please have a look at https://github.com/huggingface/diffusers/pull/254 ." - ) - - if safety_checker is not None and feature_extractor is None: - raise ValueError( - "Make sure to define a feature extractor when loading {self.__class__} if you want to use the safety" - " checker. If you do not want to use the safety checker, you can pass `'safety_checker=None'` instead." - ) - - self.register_modules( - vae=vae, - text_encoder=text_encoder, - tokenizer=tokenizer, - unet=unet, - brushnet=brushnet, - scheduler=scheduler, - safety_checker=safety_checker, - feature_extractor=feature_extractor, - image_encoder=image_encoder, - ) - self.vae_scale_factor = 2 ** (len(self.vae.config.block_out_channels) - 1) - self.image_processor = VaeImageProcessor(vae_scale_factor=self.vae_scale_factor, do_convert_rgb=True) - self.register_to_config(requires_safety_checker=requires_safety_checker) - - # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline._encode_prompt - def _encode_prompt( - self, - prompt, - device, - num_images_per_prompt, - do_classifier_free_guidance, - negative_prompt=None, - prompt_embeds: Optional[torch.FloatTensor] = None, - negative_prompt_embeds: Optional[torch.FloatTensor] = None, - lora_scale: Optional[float] = None, - **kwargs, - ): - deprecation_message = "`_encode_prompt()` is deprecated and it will be removed in a future version. Use `encode_prompt()` instead. Also, be aware that the output format changed from a concatenated tensor to a tuple." - deprecate("_encode_prompt()", "1.0.0", deprecation_message, standard_warn=False) - - prompt_embeds_tuple = self.encode_prompt( - prompt=prompt, - device=device, - num_images_per_prompt=num_images_per_prompt, - do_classifier_free_guidance=do_classifier_free_guidance, - negative_prompt=negative_prompt, - prompt_embeds=prompt_embeds, - negative_prompt_embeds=negative_prompt_embeds, - lora_scale=lora_scale, - **kwargs, - ) - - # concatenate for backwards comp - prompt_embeds = torch.cat([prompt_embeds_tuple[1], prompt_embeds_tuple[0]]) - - return prompt_embeds - - # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.encode_prompt - def encode_prompt( - self, - prompt, - device, - num_images_per_prompt, - do_classifier_free_guidance, - negative_prompt=None, - prompt_embeds: Optional[torch.FloatTensor] = None, - negative_prompt_embeds: Optional[torch.FloatTensor] = None, - lora_scale: Optional[float] = None, - clip_skip: Optional[int] = None, - ): - r""" - Encodes the prompt into text encoder hidden states. - - Args: - prompt (`str` or `List[str]`, *optional*): - prompt to be encoded - device: (`torch.device`): - torch device - num_images_per_prompt (`int`): - number of images that should be generated per prompt - do_classifier_free_guidance (`bool`): - whether to use classifier free guidance or not - negative_prompt (`str` or `List[str]`, *optional*): - The prompt or prompts not to guide the image generation. If not defined, one has to pass - `negative_prompt_embeds` instead. Ignored when not using guidance (i.e., ignored if `guidance_scale` is - less than `1`). - prompt_embeds (`torch.FloatTensor`, *optional*): - Pre-generated text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt weighting. If not - provided, text embeddings will be generated from `prompt` input argument. - negative_prompt_embeds (`torch.FloatTensor`, *optional*): - Pre-generated negative text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt - weighting. If not provided, negative_prompt_embeds will be generated from `negative_prompt` input - argument. - lora_scale (`float`, *optional*): - A LoRA scale that will be applied to all LoRA layers of the text encoder if LoRA layers are loaded. - clip_skip (`int`, *optional*): - Number of layers to be skipped from CLIP while computing the prompt embeddings. A value of 1 means that - the output of the pre-final layer will be used for computing the prompt embeddings. - """ - # set lora scale so that monkey patched LoRA - # function of text encoder can correctly access it - if lora_scale is not None and isinstance(self, LoraLoaderMixin): - self._lora_scale = lora_scale - - # dynamically adjust the LoRA scale - if not USE_PEFT_BACKEND: - adjust_lora_scale_text_encoder(self.text_encoder, lora_scale) - else: - scale_lora_layers(self.text_encoder, lora_scale) - - if prompt is not None and isinstance(prompt, str): - batch_size = 1 - elif prompt is not None and isinstance(prompt, list): - batch_size = len(prompt) - else: - batch_size = prompt_embeds.shape[0] - - if prompt_embeds is None: - # textual inversion: process multi-vector tokens if necessary - if isinstance(self, TextualInversionLoaderMixin): - prompt = self.maybe_convert_prompt(prompt, self.tokenizer) - - text_inputs = self.tokenizer( - prompt, - padding="max_length", - max_length=self.tokenizer.model_max_length, - truncation=True, - return_tensors="pt", - ) - text_input_ids = text_inputs.input_ids - untruncated_ids = self.tokenizer(prompt, padding="longest", return_tensors="pt").input_ids - - if untruncated_ids.shape[-1] >= text_input_ids.shape[-1] and not torch.equal( - text_input_ids, untruncated_ids - ): - removed_text = self.tokenizer.batch_decode( - untruncated_ids[:, self.tokenizer.model_max_length - 1 : -1] - ) - logger.warning( - "The following part of your input was truncated because CLIP can only handle sequences up to" - f" {self.tokenizer.model_max_length} tokens: {removed_text}" - ) - - if hasattr(self.text_encoder.config, "use_attention_mask") and self.text_encoder.config.use_attention_mask: - attention_mask = text_inputs.attention_mask.to(device) - else: - attention_mask = None - - if clip_skip is None: - prompt_embeds = self.text_encoder(text_input_ids.to(device), attention_mask=attention_mask) - prompt_embeds = prompt_embeds[0] - else: - prompt_embeds = self.text_encoder( - text_input_ids.to(device), attention_mask=attention_mask, output_hidden_states=True - ) - # Access the `hidden_states` first, that contains a tuple of - # all the hidden states from the encoder layers. Then index into - # the tuple to access the hidden states from the desired layer. - prompt_embeds = prompt_embeds[-1][-(clip_skip + 1)] - # We also need to apply the final LayerNorm here to not mess with the - # representations. The `last_hidden_states` that we typically use for - # obtaining the final prompt representations passes through the LayerNorm - # layer. - prompt_embeds = self.text_encoder.text_model.final_layer_norm(prompt_embeds) - - if self.text_encoder is not None: - prompt_embeds_dtype = self.text_encoder.dtype - elif self.unet is not None: - prompt_embeds_dtype = self.unet.dtype - else: - prompt_embeds_dtype = prompt_embeds.dtype - - prompt_embeds = prompt_embeds.to(dtype=prompt_embeds_dtype, device=device) - - bs_embed, seq_len, _ = prompt_embeds.shape - # duplicate text embeddings for each generation per prompt, using mps friendly method - prompt_embeds = prompt_embeds.repeat(1, num_images_per_prompt, 1) - prompt_embeds = prompt_embeds.view(bs_embed * num_images_per_prompt, seq_len, -1) - - # get unconditional embeddings for classifier free guidance - if do_classifier_free_guidance and negative_prompt_embeds is None: - uncond_tokens: List[str] - if negative_prompt is None: - uncond_tokens = [""] * batch_size - elif prompt is not None and type(prompt) is not type(negative_prompt): - raise TypeError( - f"`negative_prompt` should be the same type to `prompt`, but got {type(negative_prompt)} !=" - f" {type(prompt)}." - ) - elif isinstance(negative_prompt, str): - uncond_tokens = [negative_prompt] - elif batch_size != len(negative_prompt): - raise ValueError( - f"`negative_prompt`: {negative_prompt} has batch size {len(negative_prompt)}, but `prompt`:" - f" {prompt} has batch size {batch_size}. Please make sure that passed `negative_prompt` matches" - " the batch size of `prompt`." - ) - else: - uncond_tokens = negative_prompt - - # textual inversion: process multi-vector tokens if necessary - if isinstance(self, TextualInversionLoaderMixin): - uncond_tokens = self.maybe_convert_prompt(uncond_tokens, self.tokenizer) - - max_length = prompt_embeds.shape[1] - uncond_input = self.tokenizer( - uncond_tokens, - padding="max_length", - max_length=max_length, - truncation=True, - return_tensors="pt", - ) - - if hasattr(self.text_encoder.config, "use_attention_mask") and self.text_encoder.config.use_attention_mask: - attention_mask = uncond_input.attention_mask.to(device) - else: - attention_mask = None - - negative_prompt_embeds = self.text_encoder( - uncond_input.input_ids.to(device), - attention_mask=attention_mask, - ) - negative_prompt_embeds = negative_prompt_embeds[0] - - if do_classifier_free_guidance: - # duplicate unconditional embeddings for each generation per prompt, using mps friendly method - seq_len = negative_prompt_embeds.shape[1] - - negative_prompt_embeds = negative_prompt_embeds.to(dtype=prompt_embeds_dtype, device=device) - - negative_prompt_embeds = negative_prompt_embeds.repeat(1, num_images_per_prompt, 1) - negative_prompt_embeds = negative_prompt_embeds.view(batch_size * num_images_per_prompt, seq_len, -1) - - # self.text_encoder is not used after encode_prompt() - # self.text_encoder = None (we use prompt_embeds and negative_prompt_embeds in brushnet_nodes) - # So, nullquant commented out this section. - #if isinstance(self, LoraLoaderMixin) and USE_PEFT_BACKEND: - # # Retrieve the original scale by scaling back the LoRA layers - # unscale_lora_layers(self.text_encoder, lora_scale) - - return prompt_embeds, negative_prompt_embeds - - # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.encode_image - def encode_image(self, image, device, num_images_per_prompt, output_hidden_states=None): - dtype = next(self.image_encoder.parameters()).dtype - - if not isinstance(image, torch.Tensor): - image = self.feature_extractor(image, return_tensors="pt").pixel_values - - image = image.to(device=device, dtype=dtype) - if output_hidden_states: - image_enc_hidden_states = self.image_encoder(image, output_hidden_states=True).hidden_states[-2] - image_enc_hidden_states = image_enc_hidden_states.repeat_interleave(num_images_per_prompt, dim=0) - uncond_image_enc_hidden_states = self.image_encoder( - torch.zeros_like(image), output_hidden_states=True - ).hidden_states[-2] - uncond_image_enc_hidden_states = uncond_image_enc_hidden_states.repeat_interleave( - num_images_per_prompt, dim=0 - ) - return image_enc_hidden_states, uncond_image_enc_hidden_states - else: - image_embeds = self.image_encoder(image).image_embeds - image_embeds = image_embeds.repeat_interleave(num_images_per_prompt, dim=0) - uncond_image_embeds = torch.zeros_like(image_embeds) - - return image_embeds, uncond_image_embeds - - # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.prepare_ip_adapter_image_embeds - def prepare_ip_adapter_image_embeds( - self, ip_adapter_image, ip_adapter_image_embeds, device, num_images_per_prompt, do_classifier_free_guidance - ): - if ip_adapter_image_embeds is None: - if not isinstance(ip_adapter_image, list): - ip_adapter_image = [ip_adapter_image] - - if len(ip_adapter_image) != len(self.unet.encoder_hid_proj.image_projection_layers): - raise ValueError( - f"`ip_adapter_image` must have same length as the number of IP Adapters. Got {len(ip_adapter_image)} images and {len(self.unet.encoder_hid_proj.image_projection_layers)} IP Adapters." - ) - - image_embeds = [] - for single_ip_adapter_image, image_proj_layer in zip( - ip_adapter_image, self.unet.encoder_hid_proj.image_projection_layers - ): - output_hidden_state = not isinstance(image_proj_layer, ImageProjection) - single_image_embeds, single_negative_image_embeds = self.encode_image( - single_ip_adapter_image, device, 1, output_hidden_state - ) - single_image_embeds = torch.stack([single_image_embeds] * num_images_per_prompt, dim=0) - single_negative_image_embeds = torch.stack( - [single_negative_image_embeds] * num_images_per_prompt, dim=0 - ) - - if do_classifier_free_guidance: - single_image_embeds = torch.cat([single_negative_image_embeds, single_image_embeds]) - single_image_embeds = single_image_embeds.to(device) - - image_embeds.append(single_image_embeds) - else: - repeat_dims = [1] - image_embeds = [] - for single_image_embeds in ip_adapter_image_embeds: - if do_classifier_free_guidance: - single_negative_image_embeds, single_image_embeds = single_image_embeds.chunk(2) - single_image_embeds = single_image_embeds.repeat( - num_images_per_prompt, *(repeat_dims * len(single_image_embeds.shape[1:])) - ) - single_negative_image_embeds = single_negative_image_embeds.repeat( - num_images_per_prompt, *(repeat_dims * len(single_negative_image_embeds.shape[1:])) - ) - single_image_embeds = torch.cat([single_negative_image_embeds, single_image_embeds]) - else: - single_image_embeds = single_image_embeds.repeat( - num_images_per_prompt, *(repeat_dims * len(single_image_embeds.shape[1:])) - ) - image_embeds.append(single_image_embeds) - - return image_embeds - - # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.run_safety_checker - def run_safety_checker(self, image, device, dtype): - if self.safety_checker is None: - has_nsfw_concept = None - else: - if torch.is_tensor(image): - feature_extractor_input = self.image_processor.postprocess(image, output_type="pil") - else: - feature_extractor_input = self.image_processor.numpy_to_pil(image) - safety_checker_input = self.feature_extractor(feature_extractor_input, return_tensors="pt").to(device) - image, has_nsfw_concept = self.safety_checker( - images=image, clip_input=safety_checker_input.pixel_values.to(dtype) - ) - return image, has_nsfw_concept - - # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.decode_latents - def decode_latents(self, latents): - deprecation_message = "The decode_latents method is deprecated and will be removed in 1.0.0. Please use VaeImageProcessor.postprocess(...) instead" - deprecate("decode_latents", "1.0.0", deprecation_message, standard_warn=False) - - latents = 1 / self.vae.config.scaling_factor * latents - image = self.vae.decode(latents, return_dict=False)[0] - image = (image / 2 + 0.5).clamp(0, 1) - # we always cast to float32 as this does not cause significant overhead and is compatible with bfloat16 - image = image.cpu().permute(0, 2, 3, 1).float().numpy() - return image - - # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.prepare_extra_step_kwargs - def prepare_extra_step_kwargs(self, generator, eta): - # prepare extra kwargs for the scheduler step, since not all schedulers have the same signature - # eta (η) is only used with the DDIMScheduler, it will be ignored for other schedulers. - # eta corresponds to η in DDIM paper: https://arxiv.org/abs/2010.02502 - # and should be between [0, 1] - - accepts_eta = "eta" in set(inspect.signature(self.scheduler.step).parameters.keys()) - extra_step_kwargs = {} - if accepts_eta: - extra_step_kwargs["eta"] = eta - - # check if the scheduler accepts generator - accepts_generator = "generator" in set(inspect.signature(self.scheduler.step).parameters.keys()) - if accepts_generator: - extra_step_kwargs["generator"] = generator - return extra_step_kwargs - - def check_inputs( - self, - prompt, - image, - mask, - callback_steps, - negative_prompt=None, - prompt_embeds=None, - negative_prompt_embeds=None, - ip_adapter_image=None, - ip_adapter_image_embeds=None, - brushnet_conditioning_scale=1.0, - control_guidance_start=0.0, - control_guidance_end=1.0, - callback_on_step_end_tensor_inputs=None, - ): - if callback_steps is not None and (not isinstance(callback_steps, int) or callback_steps <= 0): - raise ValueError( - f"`callback_steps` has to be a positive integer but is {callback_steps} of type" - f" {type(callback_steps)}." - ) - - if callback_on_step_end_tensor_inputs is not None and not all( - k in self._callback_tensor_inputs for k in callback_on_step_end_tensor_inputs - ): - raise ValueError( - f"`callback_on_step_end_tensor_inputs` has to be in {self._callback_tensor_inputs}, but found {[k for k in callback_on_step_end_tensor_inputs if k not in self._callback_tensor_inputs]}" - ) - - if prompt is not None and prompt_embeds is not None: - raise ValueError( - f"Cannot forward both `prompt`: {prompt} and `prompt_embeds`: {prompt_embeds}. Please make sure to" - " only forward one of the two." - ) - elif prompt is None and prompt_embeds is None: - raise ValueError( - "Provide either `prompt` or `prompt_embeds`. Cannot leave both `prompt` and `prompt_embeds` undefined." - ) - elif prompt is not None and (not isinstance(prompt, str) and not isinstance(prompt, list)): - raise ValueError(f"`prompt` has to be of type `str` or `list` but is {type(prompt)}") - - if negative_prompt is not None and negative_prompt_embeds is not None: - raise ValueError( - f"Cannot forward both `negative_prompt`: {negative_prompt} and `negative_prompt_embeds`:" - f" {negative_prompt_embeds}. Please make sure to only forward one of the two." - ) - - if prompt_embeds is not None and negative_prompt_embeds is not None: - if prompt_embeds.shape != negative_prompt_embeds.shape: - raise ValueError( - "`prompt_embeds` and `negative_prompt_embeds` must have the same shape when passed directly, but" - f" got: `prompt_embeds` {prompt_embeds.shape} != `negative_prompt_embeds`" - f" {negative_prompt_embeds.shape}." - ) - - # Check `image` - is_compiled = hasattr(F, "scaled_dot_product_attention") and isinstance( - self.brushnet, torch._dynamo.eval_frame.OptimizedModule - ) - if ( - isinstance(self.brushnet, BrushNetModel) - or is_compiled - and isinstance(self.brushnet._orig_mod, BrushNetModel) - ): - self.check_image(image, mask, prompt, prompt_embeds) - else: - assert False - - # Check `brushnet_conditioning_scale` - if ( - isinstance(self.brushnet, BrushNetModel) - or is_compiled - and isinstance(self.brushnet._orig_mod, BrushNetModel) - ): - if not isinstance(brushnet_conditioning_scale, float): - raise TypeError("For single brushnet: `brushnet_conditioning_scale` must be type `float`.") - else: - assert False - - if not isinstance(control_guidance_start, (tuple, list)): - control_guidance_start = [control_guidance_start] - - if not isinstance(control_guidance_end, (tuple, list)): - control_guidance_end = [control_guidance_end] - - if len(control_guidance_start) != len(control_guidance_end): - raise ValueError( - f"`control_guidance_start` has {len(control_guidance_start)} elements, but `control_guidance_end` has {len(control_guidance_end)} elements. Make sure to provide the same number of elements to each list." - ) - - for start, end in zip(control_guidance_start, control_guidance_end): - if start >= end: - raise ValueError( - f"control guidance start: {start} cannot be larger or equal to control guidance end: {end}." - ) - if start < 0.0: - raise ValueError(f"control guidance start: {start} can't be smaller than 0.") - if end > 1.0: - raise ValueError(f"control guidance end: {end} can't be larger than 1.0.") - - if ip_adapter_image is not None and ip_adapter_image_embeds is not None: - raise ValueError( - "Provide either `ip_adapter_image` or `ip_adapter_image_embeds`. Cannot leave both `ip_adapter_image` and `ip_adapter_image_embeds` defined." - ) - - if ip_adapter_image_embeds is not None: - if not isinstance(ip_adapter_image_embeds, list): - raise ValueError( - f"`ip_adapter_image_embeds` has to be of type `list` but is {type(ip_adapter_image_embeds)}" - ) - elif ip_adapter_image_embeds[0].ndim not in [3, 4]: - raise ValueError( - f"`ip_adapter_image_embeds` has to be a list of 3D or 4D tensors but is {ip_adapter_image_embeds[0].ndim}D" - ) - - def check_image(self, image, mask, prompt, prompt_embeds): - image_is_pil = isinstance(image, PIL.Image.Image) - image_is_tensor = isinstance(image, torch.Tensor) - image_is_np = isinstance(image, np.ndarray) - image_is_pil_list = isinstance(image, list) and isinstance(image[0], PIL.Image.Image) - image_is_tensor_list = isinstance(image, list) and isinstance(image[0], torch.Tensor) - image_is_np_list = isinstance(image, list) and isinstance(image[0], np.ndarray) - - if ( - not image_is_pil - and not image_is_tensor - and not image_is_np - and not image_is_pil_list - and not image_is_tensor_list - and not image_is_np_list - ): - raise TypeError( - f"image must be passed and be one of PIL image, numpy array, torch tensor, list of PIL images, list of numpy arrays or list of torch tensors, but is {type(image)}" - ) - - mask_is_pil = isinstance(mask, PIL.Image.Image) - mask_is_tensor = isinstance(mask, torch.Tensor) - mask_is_np = isinstance(mask, np.ndarray) - mask_is_pil_list = isinstance(mask, list) and isinstance(mask[0], PIL.Image.Image) - mask_is_tensor_list = isinstance(mask, list) and isinstance(mask[0], torch.Tensor) - mask_is_np_list = isinstance(mask, list) and isinstance(mask[0], np.ndarray) - - if ( - not mask_is_pil - and not mask_is_tensor - and not mask_is_np - and not mask_is_pil_list - and not mask_is_tensor_list - and not mask_is_np_list - ): - raise TypeError( - f"mask must be passed and be one of PIL image, numpy array, torch tensor, list of PIL images, list of numpy arrays or list of torch tensors, but is {type(mask)}" - ) - - if image_is_pil: - image_batch_size = 1 - else: - image_batch_size = len(image) - - if prompt is not None and isinstance(prompt, str): - prompt_batch_size = 1 - elif prompt is not None and isinstance(prompt, list): - prompt_batch_size = len(prompt) - elif prompt_embeds is not None: - prompt_batch_size = prompt_embeds.shape[0] - - if image_batch_size != 1 and image_batch_size != prompt_batch_size: - raise ValueError( - f"If image batch size is not 1, image batch size must be same as prompt batch size. image batch size: {image_batch_size}, prompt batch size: {prompt_batch_size}" - ) - - def prepare_image( - self, - image, - width, - height, - batch_size, - num_images_per_prompt, - device, - dtype, - do_classifier_free_guidance=False, - guess_mode=False, - ): - image = self.image_processor.preprocess(image, height=height, width=width).to(dtype=torch.float32) - image_batch_size = image.shape[0] - - if image_batch_size == 1: - repeat_by = batch_size - else: - # image batch size is the same as prompt batch size - repeat_by = num_images_per_prompt - - image = image.repeat_interleave(repeat_by, dim=0) - - image = image.to(device=device, dtype=dtype) - - if do_classifier_free_guidance and not guess_mode: - image = torch.cat([image] * 2) - - return image - - # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.prepare_latents - def prepare_latents(self, batch_size, num_channels_latents, height, width, dtype, device, generator, latents=None): - shape = (batch_size, num_channels_latents, height // self.vae_scale_factor, width // self.vae_scale_factor) - if isinstance(generator, list) and len(generator) != batch_size: - raise ValueError( - f"You have passed a list of generators of length {len(generator)}, but requested an effective batch" - f" size of {batch_size}. Make sure the batch size matches the length of the generators." - ) - - if latents is None: - noise = randn_tensor(shape, generator=generator, device=device, dtype=dtype) - else: - noise = latents.to(device) - - # scale the initial noise by the standard deviation required by the scheduler - latents = noise * self.scheduler.init_noise_sigma - return latents, noise - - # Copied from diffusers.pipelines.latent_consistency_models.pipeline_latent_consistency_text2img.LatentConsistencyModelPipeline.get_guidance_scale_embedding - def get_guidance_scale_embedding(self, w, embedding_dim=512, dtype=torch.float32): - """ - See https://github.com/google-research/vdm/blob/dc27b98a554f65cdc654b800da5aa1846545d41b/model_vdm.py#L298 - - Args: - timesteps (`torch.Tensor`): - generate embedding vectors at these timesteps - embedding_dim (`int`, *optional*, defaults to 512): - dimension of the embeddings to generate - dtype: - data type of the generated embeddings - - Returns: - `torch.FloatTensor`: Embedding vectors with shape `(len(timesteps), embedding_dim)` - """ - assert len(w.shape) == 1 - w = w * 1000.0 - - half_dim = embedding_dim // 2 - emb = torch.log(torch.tensor(10000.0)) / (half_dim - 1) - emb = torch.exp(torch.arange(half_dim, dtype=dtype) * -emb) - emb = w.to(dtype)[:, None] * emb[None, :] - emb = torch.cat([torch.sin(emb), torch.cos(emb)], dim=1) - if embedding_dim % 2 == 1: # zero pad - emb = torch.nn.functional.pad(emb, (0, 1)) - assert emb.shape == (w.shape[0], embedding_dim) - return emb - - @property - def guidance_scale(self): - return self._guidance_scale - - @property - def clip_skip(self): - return self._clip_skip - - # here `guidance_scale` is defined analog to the guidance weight `w` of equation (2) - # of the Imagen paper: https://arxiv.org/pdf/2205.11487.pdf . `guidance_scale = 1` - # corresponds to doing no classifier free guidance. - @property - def do_classifier_free_guidance(self): - return self._guidance_scale > 1 and self.unet.config.time_cond_proj_dim is None - - @property - def cross_attention_kwargs(self): - return self._cross_attention_kwargs - - @property - def num_timesteps(self): - return self._num_timesteps - - @torch.no_grad() - @replace_example_docstring(EXAMPLE_DOC_STRING) - def __call__( - self, - prompt: Union[str, List[str]] = None, - image: PipelineImageInput = None, - mask: PipelineImageInput = None, - height: Optional[int] = None, - width: Optional[int] = None, - num_inference_steps: int = 50, - timesteps: List[int] = None, - guidance_scale: float = 7.5, - negative_prompt: Optional[Union[str, List[str]]] = None, - num_images_per_prompt: Optional[int] = 1, - eta: float = 0.0, - generator: Optional[Union[torch.Generator, List[torch.Generator]]] = None, - latents: Optional[torch.FloatTensor] = None, - prompt_embeds: Optional[torch.FloatTensor] = None, - negative_prompt_embeds: Optional[torch.FloatTensor] = None, - ip_adapter_image: Optional[PipelineImageInput] = None, - ip_adapter_image_embeds: Optional[List[torch.FloatTensor]] = None, - output_type: Optional[str] = "pil", - return_dict: bool = True, - cross_attention_kwargs: Optional[Dict[str, Any]] = None, - brushnet_conditioning_scale: Union[float, List[float]] = 1.0, - guess_mode: bool = False, - control_guidance_start: Union[float, List[float]] = 0.0, - control_guidance_end: Union[float, List[float]] = 1.0, - clip_skip: Optional[int] = None, - callback_on_step_end: Optional[Callable[[int, int, Dict], None]] = None, - callback_on_step_end_tensor_inputs: List[str] = ["latents"], - **kwargs, - ): - r""" - The call function to the pipeline for generation. - - Args: - prompt (`str` or `List[str]`, *optional*): - The prompt or prompts to guide image generation. If not defined, you need to pass `prompt_embeds`. - image (`torch.FloatTensor`, `PIL.Image.Image`, `np.ndarray`, `List[torch.FloatTensor]`, `List[PIL.Image.Image]`, `List[np.ndarray]`,: - `List[List[torch.FloatTensor]]`, `List[List[np.ndarray]]` or `List[List[PIL.Image.Image]]`): - The BrushNet input condition to provide guidance to the `unet` for generation. If the type is - specified as `torch.FloatTensor`, it is passed to BrushNet as is. `PIL.Image.Image` can also be - accepted as an image. The dimensions of the output image defaults to `image`'s dimensions. If height - and/or width are passed, `image` is resized accordingly. If multiple BrushNets are specified in - `init`, images must be passed as a list such that each element of the list can be correctly batched for - input to a single BrushNet. When `prompt` is a list, and if a list of images is passed for a single BrushNet, - each will be paired with each prompt in the `prompt` list. This also applies to multiple BrushNets, - where a list of image lists can be passed to batch for each prompt and each BrushNet. - mask (`torch.FloatTensor`, `PIL.Image.Image`, `np.ndarray`, `List[torch.FloatTensor]`, `List[PIL.Image.Image]`, `List[np.ndarray]`,: - `List[List[torch.FloatTensor]]`, `List[List[np.ndarray]]` or `List[List[PIL.Image.Image]]`): - The BrushNet input condition to provide guidance to the `unet` for generation. If the type is - specified as `torch.FloatTensor`, it is passed to BrushNet as is. `PIL.Image.Image` can also be - accepted as an image. The dimensions of the output image defaults to `image`'s dimensions. If height - and/or width are passed, `image` is resized accordingly. If multiple BrushNets are specified in - `init`, images must be passed as a list such that each element of the list can be correctly batched for - input to a single BrushNet. When `prompt` is a list, and if a list of images is passed for a single BrushNet, - each will be paired with each prompt in the `prompt` list. This also applies to multiple BrushNets, - where a list of image lists can be passed to batch for each prompt and each BrushNet. - height (`int`, *optional*, defaults to `self.unet.config.sample_size * self.vae_scale_factor`): - The height in pixels of the generated image. - width (`int`, *optional*, defaults to `self.unet.config.sample_size * self.vae_scale_factor`): - The width in pixels of the generated image. - num_inference_steps (`int`, *optional*, defaults to 50): - The number of denoising steps. More denoising steps usually lead to a higher quality image at the - expense of slower inference. - timesteps (`List[int]`, *optional*): - Custom timesteps to use for the denoising process with schedulers which support a `timesteps` argument - in their `set_timesteps` method. If not defined, the default behavior when `num_inference_steps` is - passed will be used. Must be in descending order. - guidance_scale (`float`, *optional*, defaults to 7.5): - A higher guidance scale value encourages the model to generate images closely linked to the text - `prompt` at the expense of lower image quality. Guidance scale is enabled when `guidance_scale > 1`. - negative_prompt (`str` or `List[str]`, *optional*): - The prompt or prompts to guide what to not include in image generation. If not defined, you need to - pass `negative_prompt_embeds` instead. Ignored when not using guidance (`guidance_scale < 1`). - num_images_per_prompt (`int`, *optional*, defaults to 1): - The number of images to generate per prompt. - eta (`float`, *optional*, defaults to 0.0): - Corresponds to parameter eta (η) from the [DDIM](https://arxiv.org/abs/2010.02502) paper. Only applies - to the [`~schedulers.DDIMScheduler`], and is ignored in other schedulers. - generator (`torch.Generator` or `List[torch.Generator]`, *optional*): - A [`torch.Generator`](https://pytorch.org/docs/stable/generated/torch.Generator.html) to make - generation deterministic. - latents (`torch.FloatTensor`, *optional*): - Pre-generated noisy latents sampled from a Gaussian distribution, to be used as inputs for image - generation. Can be used to tweak the same generation with different prompts. If not provided, a latents - tensor is generated by sampling using the supplied random `generator`. - prompt_embeds (`torch.FloatTensor`, *optional*): - Pre-generated text embeddings. Can be used to easily tweak text inputs (prompt weighting). If not - provided, text embeddings are generated from the `prompt` input argument. - negative_prompt_embeds (`torch.FloatTensor`, *optional*): - Pre-generated negative text embeddings. Can be used to easily tweak text inputs (prompt weighting). If - not provided, `negative_prompt_embeds` are generated from the `negative_prompt` input argument. - ip_adapter_image: (`PipelineImageInput`, *optional*): Optional image input to work with IP Adapters. - ip_adapter_image_embeds (`List[torch.FloatTensor]`, *optional*): - Pre-generated image embeddings for IP-Adapter. It should be a list of length same as number of IP-adapters. - Each element should be a tensor of shape `(batch_size, num_images, emb_dim)`. It should contain the negative image embedding - if `do_classifier_free_guidance` is set to `True`. - If not provided, embeddings are computed from the `ip_adapter_image` input argument. - output_type (`str`, *optional*, defaults to `"pil"`): - The output format of the generated image. Choose between `PIL.Image` or `np.array`. - return_dict (`bool`, *optional*, defaults to `True`): - Whether or not to return a [`~pipelines.stable_diffusion.StableDiffusionPipelineOutput`] instead of a - plain tuple. - callback (`Callable`, *optional*): - A function that calls every `callback_steps` steps during inference. The function is called with the - following arguments: `callback(step: int, timestep: int, latents: torch.FloatTensor)`. - callback_steps (`int`, *optional*, defaults to 1): - The frequency at which the `callback` function is called. If not specified, the callback is called at - every step. - cross_attention_kwargs (`dict`, *optional*): - A kwargs dictionary that if specified is passed along to the [`AttentionProcessor`] as defined in - [`self.processor`](https://github.com/huggingface/diffusers/blob/main/src/diffusers/models/attention_processor.py). - brushnet_conditioning_scale (`float` or `List[float]`, *optional*, defaults to 1.0): - The outputs of the BrushNet are multiplied by `brushnet_conditioning_scale` before they are added - to the residual in the original `unet`. If multiple BrushNets are specified in `init`, you can set - the corresponding scale as a list. - guess_mode (`bool`, *optional*, defaults to `False`): - The BrushNet encoder tries to recognize the content of the input image even if you remove all - prompts. A `guidance_scale` value between 3.0 and 5.0 is recommended. - control_guidance_start (`float` or `List[float]`, *optional*, defaults to 0.0): - The percentage of total steps at which the BrushNet starts applying. - control_guidance_end (`float` or `List[float]`, *optional*, defaults to 1.0): - The percentage of total steps at which the BrushNet stops applying. - clip_skip (`int`, *optional*): - Number of layers to be skipped from CLIP while computing the prompt embeddings. A value of 1 means that - the output of the pre-final layer will be used for computing the prompt embeddings. - callback_on_step_end (`Callable`, *optional*): - A function that calls at the end of each denoising steps during the inference. The function is called - with the following arguments: `callback_on_step_end(self: DiffusionPipeline, step: int, timestep: int, - callback_kwargs: Dict)`. `callback_kwargs` will include a list of all tensors as specified by - `callback_on_step_end_tensor_inputs`. - callback_on_step_end_tensor_inputs (`List`, *optional*): - The list of tensor inputs for the `callback_on_step_end` function. The tensors specified in the list - will be passed as `callback_kwargs` argument. You will only be able to include variables listed in the - `._callback_tensor_inputs` attribute of your pipeine class. - - Examples: - - Returns: - [`~pipelines.stable_diffusion.StableDiffusionPipelineOutput`] or `tuple`: - If `return_dict` is `True`, [`~pipelines.stable_diffusion.StableDiffusionPipelineOutput`] is returned, - otherwise a `tuple` is returned where the first element is a list with the generated images and the - second element is a list of `bool`s indicating whether the corresponding generated image contains - "not-safe-for-work" (nsfw) content. - """ - - callback = kwargs.pop("callback", None) - callback_steps = kwargs.pop("callback_steps", None) - - if callback is not None: - deprecate( - "callback", - "1.0.0", - "Passing `callback` as an input argument to `__call__` is deprecated, consider using `callback_on_step_end`", - ) - if callback_steps is not None: - deprecate( - "callback_steps", - "1.0.0", - "Passing `callback_steps` as an input argument to `__call__` is deprecated, consider using `callback_on_step_end`", - ) - - brushnet = self.brushnet._orig_mod if is_compiled_module(self.brushnet) else self.brushnet - - # align format for control guidance - if not isinstance(control_guidance_start, list) and isinstance(control_guidance_end, list): - control_guidance_start = len(control_guidance_end) * [control_guidance_start] - elif not isinstance(control_guidance_end, list) and isinstance(control_guidance_start, list): - control_guidance_end = len(control_guidance_start) * [control_guidance_end] - elif not isinstance(control_guidance_start, list) and not isinstance(control_guidance_end, list): - control_guidance_start, control_guidance_end = ( - [control_guidance_start], - [control_guidance_end], - ) - - # 1. Check inputs. Raise error if not correct - self.check_inputs( - prompt, - image, - mask, - callback_steps, - negative_prompt, - prompt_embeds, - negative_prompt_embeds, - ip_adapter_image, - ip_adapter_image_embeds, - brushnet_conditioning_scale, - control_guidance_start, - control_guidance_end, - callback_on_step_end_tensor_inputs, - ) - - self._guidance_scale = guidance_scale - self._clip_skip = clip_skip - self._cross_attention_kwargs = cross_attention_kwargs - - # 2. Define call parameters - if prompt is not None and isinstance(prompt, str): - batch_size = 1 - elif prompt is not None and isinstance(prompt, list): - batch_size = len(prompt) - else: - batch_size = prompt_embeds.shape[0] - - device = self._execution_device - - global_pool_conditions = ( - brushnet.config.global_pool_conditions - if isinstance(brushnet, BrushNetModel) - else brushnet.nets[0].config.global_pool_conditions - ) - guess_mode = guess_mode or global_pool_conditions - - # 3. Encode input prompt - text_encoder_lora_scale = ( - self.cross_attention_kwargs.get("scale", None) if self.cross_attention_kwargs is not None else None - ) - prompt_embeds, negative_prompt_embeds = self.encode_prompt( - prompt, - device, - num_images_per_prompt, - self.do_classifier_free_guidance, - negative_prompt, - prompt_embeds=prompt_embeds, - negative_prompt_embeds=negative_prompt_embeds, - lora_scale=text_encoder_lora_scale, - clip_skip=self.clip_skip, - ) - # For classifier free guidance, we need to do two forward passes. - # Here we concatenate the unconditional and text embeddings into a single batch - # to avoid doing two forward passes - if self.do_classifier_free_guidance: - prompt_embeds = torch.cat([negative_prompt_embeds, prompt_embeds]) - - if ip_adapter_image is not None or ip_adapter_image_embeds is not None: - image_embeds = self.prepare_ip_adapter_image_embeds( - ip_adapter_image, - ip_adapter_image_embeds, - device, - batch_size * num_images_per_prompt, - self.do_classifier_free_guidance, - ) - - # 4. Prepare image - if isinstance(brushnet, BrushNetModel): - image = self.prepare_image( - image=image, - width=width, - height=height, - batch_size=batch_size * num_images_per_prompt, - num_images_per_prompt=num_images_per_prompt, - device=device, - dtype=brushnet.dtype, - do_classifier_free_guidance=self.do_classifier_free_guidance, - guess_mode=guess_mode, - ) - original_mask = self.prepare_image( - image=mask, - width=width, - height=height, - batch_size=batch_size * num_images_per_prompt, - num_images_per_prompt=num_images_per_prompt, - device=device, - dtype=brushnet.dtype, - do_classifier_free_guidance=self.do_classifier_free_guidance, - guess_mode=guess_mode, - ) - original_mask=(original_mask.sum(1)[:,None,:,:] < 0).to(image.dtype) - height, width = image.shape[-2:] - else: - assert False - - # 5. Prepare timesteps - timesteps, num_inference_steps = retrieve_timesteps(self.scheduler, num_inference_steps, device, timesteps) - self._num_timesteps = len(timesteps) - - # 6. Prepare latent variables - num_channels_latents = self.unet.config.in_channels - latents, noise = self.prepare_latents( - batch_size * num_images_per_prompt, - num_channels_latents, - height, - width, - prompt_embeds.dtype, - device, - generator, - latents, - ) - - # 6.1 prepare condition latents - - conditioning_latents=self.vae.encode(image).latent_dist.sample() * self.vae.config.scaling_factor - mask = torch.nn.functional.interpolate( - original_mask, - size=( - conditioning_latents.shape[-2], - conditioning_latents.shape[-1] - ) - ) - conditioning_latents = torch.concat([conditioning_latents,mask],1) - - # 6.5 Optionally get Guidance Scale Embedding - timestep_cond = None - if self.unet.config.time_cond_proj_dim is not None: - guidance_scale_tensor = torch.tensor(self.guidance_scale - 1).repeat(batch_size * num_images_per_prompt) - timestep_cond = self.get_guidance_scale_embedding( - guidance_scale_tensor, embedding_dim=self.unet.config.time_cond_proj_dim - ).to(device=device, dtype=latents.dtype) - - # 7. Prepare extra step kwargs. TODO: Logic should ideally just be moved out of the pipeline - extra_step_kwargs = self.prepare_extra_step_kwargs(generator, eta) - - # 7.1 Add image embeds for IP-Adapter - added_cond_kwargs = ( - {"image_embeds": image_embeds} - if ip_adapter_image is not None or ip_adapter_image_embeds is not None - else None - ) - - # 7.2 Create tensor stating which brushnets to keep - brushnet_keep = [] - for i in range(len(timesteps)): - keeps = [ - 1.0 - float(i / len(timesteps) < s or (i + 1) / len(timesteps) > e) - for s, e in zip(control_guidance_start, control_guidance_end) - ] - brushnet_keep.append(keeps[0] if isinstance(brushnet, BrushNetModel) else keeps) - - # 8. Denoising loop - num_warmup_steps = len(timesteps) - num_inference_steps * self.scheduler.order - is_unet_compiled = is_compiled_module(self.unet) - is_brushnet_compiled = is_compiled_module(self.brushnet) - is_torch_higher_equal_2_1 = is_torch_version(">=", "2.1") - with self.progress_bar(total=num_inference_steps) as progress_bar: - for i, t in enumerate(timesteps): - # Relevant thread: - # https://dev-discuss.pytorch.org/t/cudagraphs-in-pytorch-2-0/1428 - if (is_unet_compiled and is_brushnet_compiled) and is_torch_higher_equal_2_1: - torch._inductor.cudagraph_mark_step_begin() - # expand the latents if we are doing classifier free guidance - latent_model_input = torch.cat([latents] * 2) if self.do_classifier_free_guidance else latents - latent_model_input = self.scheduler.scale_model_input(latent_model_input, t) - - # brushnet(s) inference - if guess_mode and self.do_classifier_free_guidance: - # Infer BrushNet only for the conditional batch. - control_model_input = latents - control_model_input = self.scheduler.scale_model_input(control_model_input, t) - brushnet_prompt_embeds = prompt_embeds.chunk(2)[1] - else: - control_model_input = latent_model_input - brushnet_prompt_embeds = prompt_embeds - - if isinstance(brushnet_keep[i], list): - cond_scale = [c * s for c, s in zip(brushnet_conditioning_scale, brushnet_keep[i])] - else: - brushnet_cond_scale = brushnet_conditioning_scale - if isinstance(brushnet_cond_scale, list): - brushnet_cond_scale = brushnet_cond_scale[0] - cond_scale = brushnet_cond_scale * brushnet_keep[i] - - down_block_res_samples, mid_block_res_sample, up_block_res_samples = self.brushnet( - control_model_input, - encoder_hidden_states=brushnet_prompt_embeds, - brushnet_cond=conditioning_latents, - timestep=t, - conditioning_scale=cond_scale, - guess_mode=guess_mode, - return_dict=False, - ) - - if guess_mode and self.do_classifier_free_guidance: - # Infered BrushNet only for the conditional batch. - # To apply the output of BrushNet to both the unconditional and conditional batches, - # add 0 to the unconditional batch to keep it unchanged. - down_block_res_samples = [torch.cat([torch.zeros_like(d), d]) for d in down_block_res_samples] - mid_block_res_sample = torch.cat([torch.zeros_like(mid_block_res_sample), mid_block_res_sample]) - up_block_res_samples = [torch.cat([torch.zeros_like(d), d]) for d in up_block_res_samples] - - # predict the noise residual - noise_pred = self.unet( - latent_model_input, - t, - encoder_hidden_states=prompt_embeds, - timestep_cond=timestep_cond, - cross_attention_kwargs=self.cross_attention_kwargs, - down_block_add_samples=down_block_res_samples, - mid_block_add_sample=mid_block_res_sample, - up_block_add_samples=up_block_res_samples, - added_cond_kwargs=added_cond_kwargs, - return_dict=False, - )[0] - - # perform guidance - if self.do_classifier_free_guidance: - noise_pred_uncond, noise_pred_text = noise_pred.chunk(2) - noise_pred = noise_pred_uncond + self.guidance_scale * (noise_pred_text - noise_pred_uncond) - - # compute the previous noisy sample x_t -> x_t-1 - latents = self.scheduler.step(noise_pred, t, latents, **extra_step_kwargs, return_dict=False)[0] - - if callback_on_step_end is not None: - callback_kwargs = {} - for k in callback_on_step_end_tensor_inputs: - callback_kwargs[k] = locals()[k] - callback_outputs = callback_on_step_end(self, i, t, callback_kwargs) - - latents = callback_outputs.pop("latents", latents) - prompt_embeds = callback_outputs.pop("prompt_embeds", prompt_embeds) - negative_prompt_embeds = callback_outputs.pop("negative_prompt_embeds", negative_prompt_embeds) - - # call the callback, if provided - if i == len(timesteps) - 1 or ((i + 1) > num_warmup_steps and (i + 1) % self.scheduler.order == 0): - progress_bar.update() - if callback is not None and i % callback_steps == 0: - step_idx = i // getattr(self.scheduler, "order", 1) - callback(step_idx, t, latents) - - # If we do sequential model offloading, let's offload unet and brushnet - # manually for max memory savings - if hasattr(self, "final_offload_hook") and self.final_offload_hook is not None: - self.unet.to("cpu") - self.brushnet.to("cpu") - torch.cuda.empty_cache() - - if not output_type == "latent": - image = self.vae.decode(latents / self.vae.config.scaling_factor, return_dict=False, generator=generator)[ - 0 - ] - image, has_nsfw_concept = self.run_safety_checker(image, device, prompt_embeds.dtype) - else: - image = latents - has_nsfw_concept = None - - if has_nsfw_concept is None: - do_denormalize = [True] * image.shape[0] - else: - do_denormalize = [not has_nsfw for has_nsfw in has_nsfw_concept] - - image = self.image_processor.postprocess(image, output_type=output_type, do_denormalize=do_denormalize) - - # Offload all models - self.maybe_free_model_hooks() - - if not return_dict: - return (image, has_nsfw_concept) - - return StableDiffusionPipelineOutput(images=image, nsfw_content_detected=has_nsfw_concept) diff --git a/brushnet/pipeline_brushnet_powerpaint.py b/brushnet/pipeline_brushnet_powerpaint.py deleted file mode 100644 index f58461a..0000000 --- a/brushnet/pipeline_brushnet_powerpaint.py +++ /dev/null @@ -1,1316 +0,0 @@ -import inspect -from typing import Any, Callable, Dict, List, Optional, Tuple, Union - -import numpy as np -import PIL.Image -import torch -import torch.nn.functional as F -from transformers import CLIPImageProcessor, CLIPTextModel, CLIPTokenizer, CLIPVisionModelWithProjection - -from ...image_processor import PipelineImageInput, VaeImageProcessor -from ...loaders import FromSingleFileMixin, IPAdapterMixin, LoraLoaderMixin, TextualInversionLoaderMixin -from ...models import AutoencoderKL, BrushNetModel, ImageProjection, UNet2DConditionModel -from ...models.lora import adjust_lora_scale_text_encoder -from ...schedulers import KarrasDiffusionSchedulers -from ...utils import ( - USE_PEFT_BACKEND, - deprecate, - logging, - replace_example_docstring, - scale_lora_layers, - unscale_lora_layers, -) -from ...utils.torch_utils import is_compiled_module, is_torch_version, randn_tensor -from ..pipeline_utils import DiffusionPipeline, StableDiffusionMixin -from ..stable_diffusion.pipeline_output import StableDiffusionPipelineOutput -from ..stable_diffusion.safety_checker import StableDiffusionSafetyChecker - - -logger = logging.get_logger(__name__) # pylint: disable=invalid-name - - -EXAMPLE_DOC_STRING = """ - Examples: - ```py - from diffusers import StableDiffusionBrushNetPipeline, BrushNetModel, UniPCMultistepScheduler - from diffusers.utils import load_image - import torch - import cv2 - import numpy as np - from PIL import Image - - base_model_path = "runwayml/stable-diffusion-v1-5" - brushnet_path = "ckpt_path" - - brushnet = BrushNetModel.from_pretrained(brushnet_path, torch_dtype=torch.float16) - pipe = StableDiffusionBrushNetPipeline.from_pretrained( - base_model_path, brushnet=brushnet, torch_dtype=torch.float16, low_cpu_mem_usage=False - ) - - # speed up diffusion process with faster scheduler and memory optimization - pipe.scheduler = UniPCMultistepScheduler.from_config(pipe.scheduler.config) - # remove following line if xformers is not installed or when using Torch 2.0. - # pipe.enable_xformers_memory_efficient_attention() - # memory optimization. - pipe.enable_model_cpu_offload() - - image_path="examples/brushnet/src/test_image.jpg" - mask_path="examples/brushnet/src/test_mask.jpg" - caption="A cake on the table." - - init_image = cv2.imread(image_path) - mask_image = 1.*(cv2.imread(mask_path).sum(-1)>255)[:,:,np.newaxis] - init_image = init_image * (1-mask_image) - - init_image = Image.fromarray(init_image.astype(np.uint8)).convert("RGB") - mask_image = Image.fromarray(mask_image.astype(np.uint8).repeat(3,-1)*255).convert("RGB") - - generator = torch.Generator("cuda").manual_seed(1234) - - image = pipe( - caption, - init_image, - mask_image, - num_inference_steps=50, - generator=generator, - paintingnet_conditioning_scale=1.0 - ).images[0] - image.save("output.png") - ``` -""" - - -# Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.retrieve_timesteps -def retrieve_timesteps( - scheduler, - num_inference_steps: Optional[int] = None, - device: Optional[Union[str, torch.device]] = None, - timesteps: Optional[List[int]] = None, - **kwargs, -): - """ - Calls the scheduler's `set_timesteps` method and retrieves timesteps from the scheduler after the call. Handles - custom timesteps. Any kwargs will be supplied to `scheduler.set_timesteps`. - - Args: - scheduler (`SchedulerMixin`): - The scheduler to get timesteps from. - num_inference_steps (`int`): - The number of diffusion steps used when generating samples with a pre-trained model. If used, - `timesteps` must be `None`. - device (`str` or `torch.device`, *optional*): - The device to which the timesteps should be moved to. If `None`, the timesteps are not moved. - timesteps (`List[int]`, *optional*): - Custom timesteps used to support arbitrary spacing between timesteps. If `None`, then the default - timestep spacing strategy of the scheduler is used. If `timesteps` is passed, `num_inference_steps` - must be `None`. - - Returns: - `Tuple[torch.Tensor, int]`: A tuple where the first element is the timestep schedule from the scheduler and the - second element is the number of inference steps. - """ - if timesteps is not None: - accepts_timesteps = "timesteps" in set(inspect.signature(scheduler.set_timesteps).parameters.keys()) - if not accepts_timesteps: - raise ValueError( - f"The current scheduler class {scheduler.__class__}'s `set_timesteps` does not support custom" - f" timestep schedules. Please check whether you are using the correct scheduler." - ) - scheduler.set_timesteps(timesteps=timesteps, device=device, **kwargs) - timesteps = scheduler.timesteps - num_inference_steps = len(timesteps) - else: - scheduler.set_timesteps(num_inference_steps, device=device, **kwargs) - timesteps = scheduler.timesteps - return timesteps, num_inference_steps - - -class StableDiffusionBrushNetPowerPaintPipeline( - DiffusionPipeline, - StableDiffusionMixin, - TextualInversionLoaderMixin, - LoraLoaderMixin, - IPAdapterMixin, - FromSingleFileMixin, -): - r""" - Pipeline for text-to-image generation using Stable Diffusion with BrushNet guidance. - - This model inherits from [`DiffusionPipeline`]. Check the superclass documentation for the generic methods - implemented for all pipelines (downloading, saving, running on a particular device, etc.). - - The pipeline also inherits the following loading methods: - - [`~loaders.TextualInversionLoaderMixin.load_textual_inversion`] for loading textual inversion embeddings - - [`~loaders.LoraLoaderMixin.load_lora_weights`] for loading LoRA weights - - [`~loaders.LoraLoaderMixin.save_lora_weights`] for saving LoRA weights - - [`~loaders.FromSingleFileMixin.from_single_file`] for loading `.ckpt` files - - [`~loaders.IPAdapterMixin.load_ip_adapter`] for loading IP Adapters - - Args: - vae ([`AutoencoderKL`]): - Variational Auto-Encoder (VAE) model to encode and decode images to and from latent representations. - text_encoder ([`~transformers.CLIPTextModel`]): - Frozen text-encoder ([clip-vit-large-patch14](https://huggingface.co/openai/clip-vit-large-patch14)). - tokenizer ([`~transformers.CLIPTokenizer`]): - A `CLIPTokenizer` to tokenize text. - unet ([`UNet2DConditionModel`]): - A `UNet2DConditionModel` to denoise the encoded image latents. - brushnet ([`BrushNetModel`]`): - Provides additional conditioning to the `unet` during the denoising process. - scheduler ([`SchedulerMixin`]): - A scheduler to be used in combination with `unet` to denoise the encoded image latents. Can be one of - [`DDIMScheduler`], [`LMSDiscreteScheduler`], or [`PNDMScheduler`]. - safety_checker ([`StableDiffusionSafetyChecker`]): - Classification module that estimates whether generated images could be considered offensive or harmful. - Please refer to the [model card](https://huggingface.co/runwayml/stable-diffusion-v1-5) for more details - about a model's potential harms. - feature_extractor ([`~transformers.CLIPImageProcessor`]): - A `CLIPImageProcessor` to extract features from generated images; used as inputs to the `safety_checker`. - """ - - model_cpu_offload_seq = "text_encoder->image_encoder->unet->vae" - _optional_components = ["safety_checker", "feature_extractor", "image_encoder"] - _exclude_from_cpu_offload = ["safety_checker"] - _callback_tensor_inputs = ["latents", "prompt_embeds", "negative_prompt_embeds"] - - def __init__( - self, - vae: AutoencoderKL, - text_encoder: CLIPTextModel, - tokenizer: CLIPTokenizer, - unet: UNet2DConditionModel, - brushnet: BrushNetModel, - scheduler: KarrasDiffusionSchedulers, - safety_checker: StableDiffusionSafetyChecker, - feature_extractor: CLIPImageProcessor, - image_encoder: CLIPVisionModelWithProjection = None, - requires_safety_checker: bool = True, - ): - super().__init__() - - if safety_checker is None and requires_safety_checker: - logger.warning( - f"You have disabled the safety checker for {self.__class__} by passing `safety_checker=None`. Ensure" - " that you abide to the conditions of the Stable Diffusion license and do not expose unfiltered" - " results in services or applications open to the public. Both the diffusers team and Hugging Face" - " strongly recommend to keep the safety filter enabled in all public facing circumstances, disabling" - " it only for use-cases that involve analyzing network behavior or auditing its results. For more" - " information, please have a look at https://github.com/huggingface/diffusers/pull/254 ." - ) - - if safety_checker is not None and feature_extractor is None: - raise ValueError( - "Make sure to define a feature extractor when loading {self.__class__} if you want to use the safety" - " checker. If you do not want to use the safety checker, you can pass `'safety_checker=None'` instead." - ) - - self.register_modules( - vae=vae, - text_encoder=text_encoder, - tokenizer=tokenizer, - unet=unet, - brushnet=brushnet, - scheduler=scheduler, - safety_checker=safety_checker, - feature_extractor=feature_extractor, - image_encoder=image_encoder, - ) - self.vae_scale_factor = 2 ** (len(self.vae.config.block_out_channels) - 1) - self.image_processor = VaeImageProcessor(vae_scale_factor=self.vae_scale_factor, do_convert_rgb=True) - self.register_to_config(requires_safety_checker=requires_safety_checker) - - # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline._encode_prompt - def _encode_prompt( - self, - prompt, - device, - num_images_per_prompt, - do_classifier_free_guidance, - negative_prompt=None, - prompt_embeds: Optional[torch.FloatTensor] = None, - negative_prompt_embeds: Optional[torch.FloatTensor] = None, - lora_scale: Optional[float] = None, - **kwargs, - ): - deprecation_message = "`_encode_prompt()` is deprecated and it will be removed in a future version. Use `encode_prompt()` instead. Also, be aware that the output format changed from a concatenated tensor to a tuple." - deprecate("_encode_prompt()", "1.0.0", deprecation_message, standard_warn=False) - - prompt_embeds_tuple = self.encode_prompt( - prompt=prompt, - device=device, - num_images_per_prompt=num_images_per_prompt, - do_classifier_free_guidance=do_classifier_free_guidance, - negative_prompt=negative_prompt, - prompt_embeds=prompt_embeds, - negative_prompt_embeds=negative_prompt_embeds, - lora_scale=lora_scale, - **kwargs, - ) - - # concatenate for backwards comp - prompt_embeds = torch.cat([prompt_embeds_tuple[1], prompt_embeds_tuple[0]]) - - return prompt_embeds - - # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.encode_prompt - def encode_prompt( - self, - prompt, - device, - num_images_per_prompt, - do_classifier_free_guidance, - negative_prompt=None, - prompt_embeds: Optional[torch.FloatTensor] = None, - negative_prompt_embeds: Optional[torch.FloatTensor] = None, - lora_scale: Optional[float] = None, - clip_skip: Optional[int] = None, - ): - r""" - Encodes the prompt into text encoder hidden states. - - Args: - prompt (`str` or `List[str]`, *optional*): - prompt to be encoded - device: (`torch.device`): - torch device - num_images_per_prompt (`int`): - number of images that should be generated per prompt - do_classifier_free_guidance (`bool`): - whether to use classifier free guidance or not - negative_prompt (`str` or `List[str]`, *optional*): - The prompt or prompts not to guide the image generation. If not defined, one has to pass - `negative_prompt_embeds` instead. Ignored when not using guidance (i.e., ignored if `guidance_scale` is - less than `1`). - prompt_embeds (`torch.FloatTensor`, *optional*): - Pre-generated text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt weighting. If not - provided, text embeddings will be generated from `prompt` input argument. - negative_prompt_embeds (`torch.FloatTensor`, *optional*): - Pre-generated negative text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt - weighting. If not provided, negative_prompt_embeds will be generated from `negative_prompt` input - argument. - lora_scale (`float`, *optional*): - A LoRA scale that will be applied to all LoRA layers of the text encoder if LoRA layers are loaded. - clip_skip (`int`, *optional*): - Number of layers to be skipped from CLIP while computing the prompt embeddings. A value of 1 means that - the output of the pre-final layer will be used for computing the prompt embeddings. - """ - # set lora scale so that monkey patched LoRA - # function of text encoder can correctly access it - print('1 ',prompt,negative_prompt) - if lora_scale is not None and isinstance(self, LoraLoaderMixin): - self._lora_scale = lora_scale - - # dynamically adjust the LoRA scale - if not USE_PEFT_BACKEND: - adjust_lora_scale_text_encoder(self.text_encoder, lora_scale) - else: - scale_lora_layers(self.text_encoder, lora_scale) - print('2 ',prompt,negative_prompt) - if prompt is not None and isinstance(prompt, str): - batch_size = 1 - elif prompt is not None and isinstance(prompt, list): - batch_size = len(prompt) - else: - batch_size = prompt_embeds.shape[0] - print('3 ',prompt,negative_prompt) - if prompt_embeds is None: - # textual inversion: process multi-vector tokens if necessary - print('4 ',prompt,negative_prompt) - if isinstance(self, TextualInversionLoaderMixin): - prompt = self.maybe_convert_prompt(prompt, self.tokenizer) - - print('5 ',prompt,negative_prompt) - - text_inputs = self.tokenizer( - prompt, - padding="max_length", - max_length=self.tokenizer.model_max_length, - truncation=True, - return_tensors="pt", - ) - text_input_ids = text_inputs.input_ids - print(prompt, text_input_ids) - untruncated_ids = self.tokenizer(prompt, padding="longest", return_tensors="pt").input_ids - - if untruncated_ids.shape[-1] >= text_input_ids.shape[-1] and not torch.equal( - text_input_ids, untruncated_ids - ): - removed_text = self.tokenizer.batch_decode( - untruncated_ids[:, self.tokenizer.model_max_length - 1 : -1] - ) - logger.warning( - "The following part of your input was truncated because CLIP can only handle sequences up to" - f" {self.tokenizer.model_max_length} tokens: {removed_text}" - ) - - if hasattr(self.text_encoder.config, "use_attention_mask") and self.text_encoder.config.use_attention_mask: - attention_mask = text_inputs.attention_mask.to(device) - else: - attention_mask = None - - if clip_skip is None: - prompt_embeds = self.text_encoder(text_input_ids.to(device), attention_mask=attention_mask) - prompt_embeds = prompt_embeds[0] - else: - prompt_embeds = self.text_encoder( - text_input_ids.to(device), attention_mask=attention_mask, output_hidden_states=True - ) - # Access the `hidden_states` first, that contains a tuple of - # all the hidden states from the encoder layers. Then index into - # the tuple to access the hidden states from the desired layer. - prompt_embeds = prompt_embeds[-1][-(clip_skip + 1)] - # We also need to apply the final LayerNorm here to not mess with the - # representations. The `last_hidden_states` that we typically use for - # obtaining the final prompt representations passes through the LayerNorm - # layer. - prompt_embeds = self.text_encoder.text_model.final_layer_norm(prompt_embeds) - - if self.text_encoder is not None: - prompt_embeds_dtype = self.text_encoder.dtype - elif self.unet is not None: - prompt_embeds_dtype = self.unet.dtype - else: - prompt_embeds_dtype = prompt_embeds.dtype - - prompt_embeds = prompt_embeds.to(dtype=prompt_embeds_dtype, device=device) - - bs_embed, seq_len, _ = prompt_embeds.shape - # duplicate text embeddings for each generation per prompt, using mps friendly method - prompt_embeds = prompt_embeds.repeat(1, num_images_per_prompt, 1) - prompt_embeds = prompt_embeds.view(bs_embed * num_images_per_prompt, seq_len, -1) - - # get unconditional embeddings for classifier free guidance - if do_classifier_free_guidance and negative_prompt_embeds is None: - uncond_tokens: List[str] - if negative_prompt is None: - uncond_tokens = [""] * batch_size - elif prompt is not None and type(prompt) is not type(negative_prompt): - raise TypeError( - f"`negative_prompt` should be the same type to `prompt`, but got {type(negative_prompt)} !=" - f" {type(prompt)}." - ) - elif isinstance(negative_prompt, str): - uncond_tokens = [negative_prompt] - elif batch_size != len(negative_prompt): - raise ValueError( - f"`negative_prompt`: {negative_prompt} has batch size {len(negative_prompt)}, but `prompt`:" - f" {prompt} has batch size {batch_size}. Please make sure that passed `negative_prompt` matches" - " the batch size of `prompt`." - ) - else: - uncond_tokens = negative_prompt - - # textual inversion: process multi-vector tokens if necessary - if isinstance(self, TextualInversionLoaderMixin): - uncond_tokens = self.maybe_convert_prompt(uncond_tokens, self.tokenizer) - - max_length = prompt_embeds.shape[1] - uncond_input = self.tokenizer( - uncond_tokens, - padding="max_length", - max_length=max_length, - truncation=True, - return_tensors="pt", - ) - print("neg: ", uncond_input.input_ids) - - if hasattr(self.text_encoder.config, "use_attention_mask") and self.text_encoder.config.use_attention_mask: - attention_mask = uncond_input.attention_mask.to(device) - else: - attention_mask = None - - negative_prompt_embeds = self.text_encoder( - uncond_input.input_ids.to(device), - attention_mask=attention_mask, - ) - negative_prompt_embeds = negative_prompt_embeds[0] - - if do_classifier_free_guidance: - # duplicate unconditional embeddings for each generation per prompt, using mps friendly method - seq_len = negative_prompt_embeds.shape[1] - - negative_prompt_embeds = negative_prompt_embeds.to(dtype=prompt_embeds_dtype, device=device) - - negative_prompt_embeds = negative_prompt_embeds.repeat(1, num_images_per_prompt, 1) - negative_prompt_embeds = negative_prompt_embeds.view(batch_size * num_images_per_prompt, seq_len, -1) - - if isinstance(self, LoraLoaderMixin) and USE_PEFT_BACKEND: - # Retrieve the original scale by scaling back the LoRA layers - unscale_lora_layers(self.text_encoder, lora_scale) - - return prompt_embeds, negative_prompt_embeds - - # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.encode_image - def encode_image(self, image, device, num_images_per_prompt, output_hidden_states=None): - dtype = next(self.image_encoder.parameters()).dtype - - if not isinstance(image, torch.Tensor): - image = self.feature_extractor(image, return_tensors="pt").pixel_values - - image = image.to(device=device, dtype=dtype) - if output_hidden_states: - image_enc_hidden_states = self.image_encoder(image, output_hidden_states=True).hidden_states[-2] - image_enc_hidden_states = image_enc_hidden_states.repeat_interleave(num_images_per_prompt, dim=0) - uncond_image_enc_hidden_states = self.image_encoder( - torch.zeros_like(image), output_hidden_states=True - ).hidden_states[-2] - uncond_image_enc_hidden_states = uncond_image_enc_hidden_states.repeat_interleave( - num_images_per_prompt, dim=0 - ) - return image_enc_hidden_states, uncond_image_enc_hidden_states - else: - image_embeds = self.image_encoder(image).image_embeds - image_embeds = image_embeds.repeat_interleave(num_images_per_prompt, dim=0) - uncond_image_embeds = torch.zeros_like(image_embeds) - - return image_embeds, uncond_image_embeds - - # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.prepare_ip_adapter_image_embeds - def prepare_ip_adapter_image_embeds( - self, ip_adapter_image, ip_adapter_image_embeds, device, num_images_per_prompt, do_classifier_free_guidance - ): - if ip_adapter_image_embeds is None: - if not isinstance(ip_adapter_image, list): - ip_adapter_image = [ip_adapter_image] - - if len(ip_adapter_image) != len(self.unet.encoder_hid_proj.image_projection_layers): - raise ValueError( - f"`ip_adapter_image` must have same length as the number of IP Adapters. Got {len(ip_adapter_image)} images and {len(self.unet.encoder_hid_proj.image_projection_layers)} IP Adapters." - ) - - image_embeds = [] - for single_ip_adapter_image, image_proj_layer in zip( - ip_adapter_image, self.unet.encoder_hid_proj.image_projection_layers - ): - output_hidden_state = not isinstance(image_proj_layer, ImageProjection) - single_image_embeds, single_negative_image_embeds = self.encode_image( - single_ip_adapter_image, device, 1, output_hidden_state - ) - single_image_embeds = torch.stack([single_image_embeds] * num_images_per_prompt, dim=0) - single_negative_image_embeds = torch.stack( - [single_negative_image_embeds] * num_images_per_prompt, dim=0 - ) - - if do_classifier_free_guidance: - single_image_embeds = torch.cat([single_negative_image_embeds, single_image_embeds]) - single_image_embeds = single_image_embeds.to(device) - - image_embeds.append(single_image_embeds) - else: - repeat_dims = [1] - image_embeds = [] - for single_image_embeds in ip_adapter_image_embeds: - if do_classifier_free_guidance: - single_negative_image_embeds, single_image_embeds = single_image_embeds.chunk(2) - single_image_embeds = single_image_embeds.repeat( - num_images_per_prompt, *(repeat_dims * len(single_image_embeds.shape[1:])) - ) - single_negative_image_embeds = single_negative_image_embeds.repeat( - num_images_per_prompt, *(repeat_dims * len(single_negative_image_embeds.shape[1:])) - ) - single_image_embeds = torch.cat([single_negative_image_embeds, single_image_embeds]) - else: - single_image_embeds = single_image_embeds.repeat( - num_images_per_prompt, *(repeat_dims * len(single_image_embeds.shape[1:])) - ) - image_embeds.append(single_image_embeds) - - return image_embeds - - # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.run_safety_checker - def run_safety_checker(self, image, device, dtype): - if self.safety_checker is None: - has_nsfw_concept = None - else: - if torch.is_tensor(image): - feature_extractor_input = self.image_processor.postprocess(image, output_type="pil") - else: - feature_extractor_input = self.image_processor.numpy_to_pil(image) - safety_checker_input = self.feature_extractor(feature_extractor_input, return_tensors="pt").to(device) - image, has_nsfw_concept = self.safety_checker( - images=image, clip_input=safety_checker_input.pixel_values.to(dtype) - ) - return image, has_nsfw_concept - - # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.decode_latents - def decode_latents(self, latents): - deprecation_message = "The decode_latents method is deprecated and will be removed in 1.0.0. Please use VaeImageProcessor.postprocess(...) instead" - deprecate("decode_latents", "1.0.0", deprecation_message, standard_warn=False) - - latents = 1 / self.vae.config.scaling_factor * latents - image = self.vae.decode(latents, return_dict=False)[0] - image = (image / 2 + 0.5).clamp(0, 1) - # we always cast to float32 as this does not cause significant overhead and is compatible with bfloat16 - image = image.cpu().permute(0, 2, 3, 1).float().numpy() - return image - - # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.prepare_extra_step_kwargs - def prepare_extra_step_kwargs(self, generator, eta): - # prepare extra kwargs for the scheduler step, since not all schedulers have the same signature - # eta (η) is only used with the DDIMScheduler, it will be ignored for other schedulers. - # eta corresponds to η in DDIM paper: https://arxiv.org/abs/2010.02502 - # and should be between [0, 1] - - accepts_eta = "eta" in set(inspect.signature(self.scheduler.step).parameters.keys()) - extra_step_kwargs = {} - if accepts_eta: - extra_step_kwargs["eta"] = eta - - # check if the scheduler accepts generator - accepts_generator = "generator" in set(inspect.signature(self.scheduler.step).parameters.keys()) - if accepts_generator: - extra_step_kwargs["generator"] = generator - return extra_step_kwargs - - def check_inputs( - self, - prompt, - image, - mask, - callback_steps, - negative_prompt=None, - prompt_embeds=None, - negative_prompt_embeds=None, - ip_adapter_image=None, - ip_adapter_image_embeds=None, - brushnet_conditioning_scale=1.0, - control_guidance_start=0.0, - control_guidance_end=1.0, - callback_on_step_end_tensor_inputs=None, - ): - if callback_steps is not None and (not isinstance(callback_steps, int) or callback_steps <= 0): - raise ValueError( - f"`callback_steps` has to be a positive integer but is {callback_steps} of type" - f" {type(callback_steps)}." - ) - - if callback_on_step_end_tensor_inputs is not None and not all( - k in self._callback_tensor_inputs for k in callback_on_step_end_tensor_inputs - ): - raise ValueError( - f"`callback_on_step_end_tensor_inputs` has to be in {self._callback_tensor_inputs}, but found {[k for k in callback_on_step_end_tensor_inputs if k not in self._callback_tensor_inputs]}" - ) - - if prompt is not None and prompt_embeds is not None: - raise ValueError( - f"Cannot forward both `prompt`: {prompt} and `prompt_embeds`: {prompt_embeds}. Please make sure to" - " only forward one of the two." - ) - elif prompt is None and prompt_embeds is None: - raise ValueError( - "Provide either `prompt` or `prompt_embeds`. Cannot leave both `prompt` and `prompt_embeds` undefined." - ) - elif prompt is not None and (not isinstance(prompt, str) and not isinstance(prompt, list)): - raise ValueError(f"`prompt` has to be of type `str` or `list` but is {type(prompt)}") - - if negative_prompt is not None and negative_prompt_embeds is not None: - raise ValueError( - f"Cannot forward both `negative_prompt`: {negative_prompt} and `negative_prompt_embeds`:" - f" {negative_prompt_embeds}. Please make sure to only forward one of the two." - ) - - if prompt_embeds is not None and negative_prompt_embeds is not None: - if prompt_embeds.shape != negative_prompt_embeds.shape: - raise ValueError( - "`prompt_embeds` and `negative_prompt_embeds` must have the same shape when passed directly, but" - f" got: `prompt_embeds` {prompt_embeds.shape} != `negative_prompt_embeds`" - f" {negative_prompt_embeds.shape}." - ) - - # Check `image` - is_compiled = hasattr(F, "scaled_dot_product_attention") and isinstance( - self.brushnet, torch._dynamo.eval_frame.OptimizedModule - ) - if ( - isinstance(self.brushnet, BrushNetModel) - or is_compiled - and isinstance(self.brushnet._orig_mod, BrushNetModel) - ): - self.check_image(image, mask, prompt, prompt_embeds) - else: - assert False - - # Check `brushnet_conditioning_scale` - if ( - isinstance(self.brushnet, BrushNetModel) - or is_compiled - and isinstance(self.brushnet._orig_mod, BrushNetModel) - ): - if not isinstance(brushnet_conditioning_scale, float): - raise TypeError("For single brushnet: `brushnet_conditioning_scale` must be type `float`.") - else: - assert False - - if not isinstance(control_guidance_start, (tuple, list)): - control_guidance_start = [control_guidance_start] - - if not isinstance(control_guidance_end, (tuple, list)): - control_guidance_end = [control_guidance_end] - - if len(control_guidance_start) != len(control_guidance_end): - raise ValueError( - f"`control_guidance_start` has {len(control_guidance_start)} elements, but `control_guidance_end` has {len(control_guidance_end)} elements. Make sure to provide the same number of elements to each list." - ) - - for start, end in zip(control_guidance_start, control_guidance_end): - if start >= end: - raise ValueError( - f"control guidance start: {start} cannot be larger or equal to control guidance end: {end}." - ) - if start < 0.0: - raise ValueError(f"control guidance start: {start} can't be smaller than 0.") - if end > 1.0: - raise ValueError(f"control guidance end: {end} can't be larger than 1.0.") - - if ip_adapter_image is not None and ip_adapter_image_embeds is not None: - raise ValueError( - "Provide either `ip_adapter_image` or `ip_adapter_image_embeds`. Cannot leave both `ip_adapter_image` and `ip_adapter_image_embeds` defined." - ) - - if ip_adapter_image_embeds is not None: - if not isinstance(ip_adapter_image_embeds, list): - raise ValueError( - f"`ip_adapter_image_embeds` has to be of type `list` but is {type(ip_adapter_image_embeds)}" - ) - elif ip_adapter_image_embeds[0].ndim not in [3, 4]: - raise ValueError( - f"`ip_adapter_image_embeds` has to be a list of 3D or 4D tensors but is {ip_adapter_image_embeds[0].ndim}D" - ) - - def check_image(self, image, mask, prompt, prompt_embeds): - image_is_pil = isinstance(image, PIL.Image.Image) - image_is_tensor = isinstance(image, torch.Tensor) - image_is_np = isinstance(image, np.ndarray) - image_is_pil_list = isinstance(image, list) and isinstance(image[0], PIL.Image.Image) - image_is_tensor_list = isinstance(image, list) and isinstance(image[0], torch.Tensor) - image_is_np_list = isinstance(image, list) and isinstance(image[0], np.ndarray) - - if ( - not image_is_pil - and not image_is_tensor - and not image_is_np - and not image_is_pil_list - and not image_is_tensor_list - and not image_is_np_list - ): - raise TypeError( - f"image must be passed and be one of PIL image, numpy array, torch tensor, list of PIL images, list of numpy arrays or list of torch tensors, but is {type(image)}" - ) - - mask_is_pil = isinstance(mask, PIL.Image.Image) - mask_is_tensor = isinstance(mask, torch.Tensor) - mask_is_np = isinstance(mask, np.ndarray) - mask_is_pil_list = isinstance(mask, list) and isinstance(mask[0], PIL.Image.Image) - mask_is_tensor_list = isinstance(mask, list) and isinstance(mask[0], torch.Tensor) - mask_is_np_list = isinstance(mask, list) and isinstance(mask[0], np.ndarray) - - if ( - not mask_is_pil - and not mask_is_tensor - and not mask_is_np - and not mask_is_pil_list - and not mask_is_tensor_list - and not mask_is_np_list - ): - raise TypeError( - f"mask must be passed and be one of PIL image, numpy array, torch tensor, list of PIL images, list of numpy arrays or list of torch tensors, but is {type(mask)}" - ) - - if image_is_pil: - image_batch_size = 1 - else: - image_batch_size = len(image) - - if prompt is not None and isinstance(prompt, str): - prompt_batch_size = 1 - elif prompt is not None and isinstance(prompt, list): - prompt_batch_size = len(prompt) - elif prompt_embeds is not None: - prompt_batch_size = prompt_embeds.shape[0] - - if image_batch_size != 1 and image_batch_size != prompt_batch_size: - raise ValueError( - f"If image batch size is not 1, image batch size must be same as prompt batch size. image batch size: {image_batch_size}, prompt batch size: {prompt_batch_size}" - ) - - def prepare_image( - self, - image, - width, - height, - batch_size, - num_images_per_prompt, - device, - dtype, - do_classifier_free_guidance=False, - guess_mode=False, - ): - image = self.image_processor.preprocess(image, height=height, width=width).to(dtype=torch.float32) - image_batch_size = image.shape[0] - - if image_batch_size == 1: - repeat_by = batch_size - else: - # image batch size is the same as prompt batch size - repeat_by = num_images_per_prompt - - image = image.repeat_interleave(repeat_by, dim=0) - - image = image.to(device=device, dtype=dtype) - - if do_classifier_free_guidance and not guess_mode: - image = torch.cat([image] * 2) - - return image - - # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.prepare_latents - def prepare_latents(self, batch_size, num_channels_latents, height, width, dtype, device, generator, latents=None): - shape = (batch_size, num_channels_latents, height // self.vae_scale_factor, width // self.vae_scale_factor) - if isinstance(generator, list) and len(generator) != batch_size: - raise ValueError( - f"You have passed a list of generators of length {len(generator)}, but requested an effective batch" - f" size of {batch_size}. Make sure the batch size matches the length of the generators." - ) - - if latents is None: - noise = randn_tensor(shape, generator=generator, device=device, dtype=dtype) - else: - noise = latents.to(device) - - # scale the initial noise by the standard deviation required by the scheduler - latents = noise * self.scheduler.init_noise_sigma - return latents, noise - - # Copied from diffusers.pipelines.latent_consistency_models.pipeline_latent_consistency_text2img.LatentConsistencyModelPipeline.get_guidance_scale_embedding - def get_guidance_scale_embedding(self, w, embedding_dim=512, dtype=torch.float32): - """ - See https://github.com/google-research/vdm/blob/dc27b98a554f65cdc654b800da5aa1846545d41b/model_vdm.py#L298 - - Args: - timesteps (`torch.Tensor`): - generate embedding vectors at these timesteps - embedding_dim (`int`, *optional*, defaults to 512): - dimension of the embeddings to generate - dtype: - data type of the generated embeddings - - Returns: - `torch.FloatTensor`: Embedding vectors with shape `(len(timesteps), embedding_dim)` - """ - assert len(w.shape) == 1 - w = w * 1000.0 - - half_dim = embedding_dim // 2 - emb = torch.log(torch.tensor(10000.0)) / (half_dim - 1) - emb = torch.exp(torch.arange(half_dim, dtype=dtype) * -emb) - emb = w.to(dtype)[:, None] * emb[None, :] - emb = torch.cat([torch.sin(emb), torch.cos(emb)], dim=1) - if embedding_dim % 2 == 1: # zero pad - emb = torch.nn.functional.pad(emb, (0, 1)) - assert emb.shape == (w.shape[0], embedding_dim) - return emb - - @property - def guidance_scale(self): - return self._guidance_scale - - @property - def clip_skip(self): - return self._clip_skip - - # here `guidance_scale` is defined analog to the guidance weight `w` of equation (2) - # of the Imagen paper: https://arxiv.org/pdf/2205.11487.pdf . `guidance_scale = 1` - # corresponds to doing no classifier free guidance. - @property - def do_classifier_free_guidance(self): - return self._guidance_scale > 1 and self.unet.config.time_cond_proj_dim is None - - @property - def cross_attention_kwargs(self): - return self._cross_attention_kwargs - - @property - def num_timesteps(self): - return self._num_timesteps - - @torch.no_grad() - @replace_example_docstring(EXAMPLE_DOC_STRING) - def __call__( - self, - prompt: Union[str, List[str]] = None, - image: PipelineImageInput = None, - mask: PipelineImageInput = None, - height: Optional[int] = None, - width: Optional[int] = None, - num_inference_steps: int = 50, - timesteps: List[int] = None, - guidance_scale: float = 7.5, - negative_prompt: Optional[Union[str, List[str]]] = None, - num_images_per_prompt: Optional[int] = 1, - eta: float = 0.0, - generator: Optional[Union[torch.Generator, List[torch.Generator]]] = None, - latents: Optional[torch.FloatTensor] = None, - prompt_embeds: Optional[torch.FloatTensor] = None, - negative_prompt_embeds: Optional[torch.FloatTensor] = None, - ip_adapter_image: Optional[PipelineImageInput] = None, - ip_adapter_image_embeds: Optional[List[torch.FloatTensor]] = None, - output_type: Optional[str] = "pil", - return_dict: bool = True, - cross_attention_kwargs: Optional[Dict[str, Any]] = None, - brushnet_conditioning_scale: Union[float, List[float]] = 1.0, - guess_mode: bool = False, - control_guidance_start: Union[float, List[float]] = 0.0, - control_guidance_end: Union[float, List[float]] = 1.0, - clip_skip: Optional[int] = None, - callback_on_step_end: Optional[Callable[[int, int, Dict], None]] = None, - callback_on_step_end_tensor_inputs: List[str] = ["latents"], - con_prompt = None, - con_negprompt = None, - **kwargs, - ): - r""" - The call function to the pipeline for generation. - - Args: - prompt (`str` or `List[str]`, *optional*): - The prompt or prompts to guide image generation. If not defined, you need to pass `prompt_embeds`. - image (`torch.FloatTensor`, `PIL.Image.Image`, `np.ndarray`, `List[torch.FloatTensor]`, `List[PIL.Image.Image]`, `List[np.ndarray]`,: - `List[List[torch.FloatTensor]]`, `List[List[np.ndarray]]` or `List[List[PIL.Image.Image]]`): - The BrushNet input condition to provide guidance to the `unet` for generation. If the type is - specified as `torch.FloatTensor`, it is passed to BrushNet as is. `PIL.Image.Image` can also be - accepted as an image. The dimensions of the output image defaults to `image`'s dimensions. If height - and/or width are passed, `image` is resized accordingly. If multiple BrushNets are specified in - `init`, images must be passed as a list such that each element of the list can be correctly batched for - input to a single BrushNet. When `prompt` is a list, and if a list of images is passed for a single BrushNet, - each will be paired with each prompt in the `prompt` list. This also applies to multiple BrushNets, - where a list of image lists can be passed to batch for each prompt and each BrushNet. - mask (`torch.FloatTensor`, `PIL.Image.Image`, `np.ndarray`, `List[torch.FloatTensor]`, `List[PIL.Image.Image]`, `List[np.ndarray]`,: - `List[List[torch.FloatTensor]]`, `List[List[np.ndarray]]` or `List[List[PIL.Image.Image]]`): - The BrushNet input condition to provide guidance to the `unet` for generation. If the type is - specified as `torch.FloatTensor`, it is passed to BrushNet as is. `PIL.Image.Image` can also be - accepted as an image. The dimensions of the output image defaults to `image`'s dimensions. If height - and/or width are passed, `image` is resized accordingly. If multiple BrushNets are specified in - `init`, images must be passed as a list such that each element of the list can be correctly batched for - input to a single BrushNet. When `prompt` is a list, and if a list of images is passed for a single BrushNet, - each will be paired with each prompt in the `prompt` list. This also applies to multiple BrushNets, - where a list of image lists can be passed to batch for each prompt and each BrushNet. - height (`int`, *optional*, defaults to `self.unet.config.sample_size * self.vae_scale_factor`): - The height in pixels of the generated image. - width (`int`, *optional*, defaults to `self.unet.config.sample_size * self.vae_scale_factor`): - The width in pixels of the generated image. - num_inference_steps (`int`, *optional*, defaults to 50): - The number of denoising steps. More denoising steps usually lead to a higher quality image at the - expense of slower inference. - timesteps (`List[int]`, *optional*): - Custom timesteps to use for the denoising process with schedulers which support a `timesteps` argument - in their `set_timesteps` method. If not defined, the default behavior when `num_inference_steps` is - passed will be used. Must be in descending order. - guidance_scale (`float`, *optional*, defaults to 7.5): - A higher guidance scale value encourages the model to generate images closely linked to the text - `prompt` at the expense of lower image quality. Guidance scale is enabled when `guidance_scale > 1`. - negative_prompt (`str` or `List[str]`, *optional*): - The prompt or prompts to guide what to not include in image generation. If not defined, you need to - pass `negative_prompt_embeds` instead. Ignored when not using guidance (`guidance_scale < 1`). - num_images_per_prompt (`int`, *optional*, defaults to 1): - The number of images to generate per prompt. - eta (`float`, *optional*, defaults to 0.0): - Corresponds to parameter eta (η) from the [DDIM](https://arxiv.org/abs/2010.02502) paper. Only applies - to the [`~schedulers.DDIMScheduler`], and is ignored in other schedulers. - generator (`torch.Generator` or `List[torch.Generator]`, *optional*): - A [`torch.Generator`](https://pytorch.org/docs/stable/generated/torch.Generator.html) to make - generation deterministic. - latents (`torch.FloatTensor`, *optional*): - Pre-generated noisy latents sampled from a Gaussian distribution, to be used as inputs for image - generation. Can be used to tweak the same generation with different prompts. If not provided, a latents - tensor is generated by sampling using the supplied random `generator`. - prompt_embeds (`torch.FloatTensor`, *optional*): - Pre-generated text embeddings. Can be used to easily tweak text inputs (prompt weighting). If not - provided, text embeddings are generated from the `prompt` input argument. - negative_prompt_embeds (`torch.FloatTensor`, *optional*): - Pre-generated negative text embeddings. Can be used to easily tweak text inputs (prompt weighting). If - not provided, `negative_prompt_embeds` are generated from the `negative_prompt` input argument. - ip_adapter_image: (`PipelineImageInput`, *optional*): Optional image input to work with IP Adapters. - ip_adapter_image_embeds (`List[torch.FloatTensor]`, *optional*): - Pre-generated image embeddings for IP-Adapter. It should be a list of length same as number of IP-adapters. - Each element should be a tensor of shape `(batch_size, num_images, emb_dim)`. It should contain the negative image embedding - if `do_classifier_free_guidance` is set to `True`. - If not provided, embeddings are computed from the `ip_adapter_image` input argument. - output_type (`str`, *optional*, defaults to `"pil"`): - The output format of the generated image. Choose between `PIL.Image` or `np.array`. - return_dict (`bool`, *optional*, defaults to `True`): - Whether or not to return a [`~pipelines.stable_diffusion.StableDiffusionPipelineOutput`] instead of a - plain tuple. - callback (`Callable`, *optional*): - A function that calls every `callback_steps` steps during inference. The function is called with the - following arguments: `callback(step: int, timestep: int, latents: torch.FloatTensor)`. - callback_steps (`int`, *optional*, defaults to 1): - The frequency at which the `callback` function is called. If not specified, the callback is called at - every step. - cross_attention_kwargs (`dict`, *optional*): - A kwargs dictionary that if specified is passed along to the [`AttentionProcessor`] as defined in - [`self.processor`](https://github.com/huggingface/diffusers/blob/main/src/diffusers/models/attention_processor.py). - brushnet_conditioning_scale (`float` or `List[float]`, *optional*, defaults to 1.0): - The outputs of the BrushNet are multiplied by `brushnet_conditioning_scale` before they are added - to the residual in the original `unet`. If multiple BrushNets are specified in `init`, you can set - the corresponding scale as a list. - guess_mode (`bool`, *optional*, defaults to `False`): - The BrushNet encoder tries to recognize the content of the input image even if you remove all - prompts. A `guidance_scale` value between 3.0 and 5.0 is recommended. - control_guidance_start (`float` or `List[float]`, *optional*, defaults to 0.0): - The percentage of total steps at which the BrushNet starts applying. - control_guidance_end (`float` or `List[float]`, *optional*, defaults to 1.0): - The percentage of total steps at which the BrushNet stops applying. - clip_skip (`int`, *optional*): - Number of layers to be skipped from CLIP while computing the prompt embeddings. A value of 1 means that - the output of the pre-final layer will be used for computing the prompt embeddings. - callback_on_step_end (`Callable`, *optional*): - A function that calls at the end of each denoising steps during the inference. The function is called - with the following arguments: `callback_on_step_end(self: DiffusionPipeline, step: int, timestep: int, - callback_kwargs: Dict)`. `callback_kwargs` will include a list of all tensors as specified by - `callback_on_step_end_tensor_inputs`. - callback_on_step_end_tensor_inputs (`List`, *optional*): - The list of tensor inputs for the `callback_on_step_end` function. The tensors specified in the list - will be passed as `callback_kwargs` argument. You will only be able to include variables listed in the - `._callback_tensor_inputs` attribute of your pipeine class. - - Examples: - - Returns: - [`~pipelines.stable_diffusion.StableDiffusionPipelineOutput`] or `tuple`: - If `return_dict` is `True`, [`~pipelines.stable_diffusion.StableDiffusionPipelineOutput`] is returned, - otherwise a `tuple` is returned where the first element is a list with the generated images and the - second element is a list of `bool`s indicating whether the corresponding generated image contains - "not-safe-for-work" (nsfw) content. - """ - - callback = kwargs.pop("callback", None) - callback_steps = kwargs.pop("callback_steps", None) - - if callback is not None: - deprecate( - "callback", - "1.0.0", - "Passing `callback` as an input argument to `__call__` is deprecated, consider using `callback_on_step_end`", - ) - if callback_steps is not None: - deprecate( - "callback_steps", - "1.0.0", - "Passing `callback_steps` as an input argument to `__call__` is deprecated, consider using `callback_on_step_end`", - ) - - brushnet = self.brushnet._orig_mod if is_compiled_module(self.brushnet) else self.brushnet - - # align format for control guidance - if not isinstance(control_guidance_start, list) and isinstance(control_guidance_end, list): - control_guidance_start = len(control_guidance_end) * [control_guidance_start] - elif not isinstance(control_guidance_end, list) and isinstance(control_guidance_start, list): - control_guidance_end = len(control_guidance_start) * [control_guidance_end] - elif not isinstance(control_guidance_start, list) and not isinstance(control_guidance_end, list): - control_guidance_start, control_guidance_end = ( - [control_guidance_start], - [control_guidance_end], - ) - - # 1. Check inputs. Raise error if not correct - self.check_inputs( - prompt, - image, - mask, - callback_steps, - negative_prompt, - prompt_embeds, - negative_prompt_embeds, - ip_adapter_image, - ip_adapter_image_embeds, - brushnet_conditioning_scale, - control_guidance_start, - control_guidance_end, - callback_on_step_end_tensor_inputs, - ) - - self._guidance_scale = guidance_scale - self._clip_skip = clip_skip - self._cross_attention_kwargs = cross_attention_kwargs - - # 2. Define call parameters - if prompt is not None and isinstance(prompt, str): - batch_size = 1 - elif prompt is not None and isinstance(prompt, list): - batch_size = len(prompt) - else: - batch_size = prompt_embeds.shape[0] - - device = self._execution_device - - global_pool_conditions = ( - brushnet.config.global_pool_conditions - if isinstance(brushnet, BrushNetModel) - else brushnet.nets[0].config.global_pool_conditions - ) - guess_mode = guess_mode or global_pool_conditions - - # 3. Encode input prompt - text_encoder_lora_scale = ( - self.cross_attention_kwargs.get("scale", None) if self.cross_attention_kwargs is not None else None - ) - prompt_embeds, negative_prompt_embeds = self.encode_prompt( - prompt, - device, - num_images_per_prompt, - self.do_classifier_free_guidance, - negative_prompt, - prompt_embeds=prompt_embeds, - negative_prompt_embeds=negative_prompt_embeds, - lora_scale=text_encoder_lora_scale, - clip_skip=self.clip_skip, - ) - - print(con_prompt, con_negprompt) - con_negative_prompt_embeds = None - con_prompt_embeds = None - con_prompt_embeds, con_negative_prompt_embeds = self.encode_prompt( - con_prompt, - device, - num_images_per_prompt, - self.do_classifier_free_guidance, - con_negprompt, - prompt_embeds=con_prompt_embeds, - negative_prompt_embeds=con_negative_prompt_embeds, - lora_scale=text_encoder_lora_scale, - clip_skip=self.clip_skip, - ) - - - # For classifier free guidance, we need to do two forward passes. - # Here we concatenate the unconditional and text embeddings into a single batch - # to avoid doing two forward passes - if self.do_classifier_free_guidance: - prompt_embeds = torch.cat([negative_prompt_embeds, prompt_embeds]) - con_prompt_embeds = torch.cat([con_negative_prompt_embeds, con_prompt_embeds]) - - if ip_adapter_image is not None or ip_adapter_image_embeds is not None: - image_embeds = self.prepare_ip_adapter_image_embeds( - ip_adapter_image, - ip_adapter_image_embeds, - device, - batch_size * num_images_per_prompt, - self.do_classifier_free_guidance, - ) - - # 4. Prepare image - if isinstance(brushnet, BrushNetModel): - image = self.prepare_image( - image=image, - width=width, - height=height, - batch_size=batch_size * num_images_per_prompt, - num_images_per_prompt=num_images_per_prompt, - device=device, - dtype=brushnet.dtype, - do_classifier_free_guidance=self.do_classifier_free_guidance, - guess_mode=guess_mode, - ) - original_mask = self.prepare_image( - image=mask, - width=width, - height=height, - batch_size=batch_size * num_images_per_prompt, - num_images_per_prompt=num_images_per_prompt, - device=device, - dtype=brushnet.dtype, - do_classifier_free_guidance=self.do_classifier_free_guidance, - guess_mode=guess_mode, - ) - original_mask=(original_mask.sum(1)[:,None,:,:] < 0).to(image.dtype) - height, width = image.shape[-2:] - else: - assert False - - # 5. Prepare timesteps - timesteps, num_inference_steps = retrieve_timesteps(self.scheduler, num_inference_steps, device, timesteps) - self._num_timesteps = len(timesteps) - - # 6. Prepare latent variables - num_channels_latents = self.unet.config.in_channels - latents, noise = self.prepare_latents( - batch_size * num_images_per_prompt, - num_channels_latents, - height, - width, - prompt_embeds.dtype, - device, - generator, - latents, - ) - - # 6.1 prepare condition latents - from torchvision import transforms - # mask_i = transforms.ToPILImage()(image[0:1,:,:,:].squeeze(0)) - # mask_i.save('_mask.png') - - print(image) - mask_i = transforms.ToPILImage()(image[0:1,:,:,:].squeeze(0)/2+0.5) - mask_i.save(str(timesteps[0]) +'_CC.png') - conditioning_latents=self.vae.encode(image).latent_dist.sample() * self.vae.config.scaling_factor - mask = torch.nn.functional.interpolate( - original_mask, - size=( - conditioning_latents.shape[-2], - conditioning_latents.shape[-1] - ) - ) - conditioning_latents = torch.concat([conditioning_latents,mask],1) - # image = self.vae.decode(conditioning_latents[:1,:4,:,:] / self.vae.config.scaling_factor, return_dict=False, generator=generator)[0] - # from torchvision import transforms - # mask_i = transforms.ToPILImage()(image[0:1,:,:,:].squeeze(0)/2+0.5) - # mask_i.save(str(timesteps[0]) +'_C.png') - - - # 6.5 Optionally get Guidance Scale Embedding - timestep_cond = None - if self.unet.config.time_cond_proj_dim is not None: - guidance_scale_tensor = torch.tensor(self.guidance_scale - 1).repeat(batch_size * num_images_per_prompt) - timestep_cond = self.get_guidance_scale_embedding( - guidance_scale_tensor, embedding_dim=self.unet.config.time_cond_proj_dim - ).to(device=device, dtype=latents.dtype) - - # 7. Prepare extra step kwargs. TODO: Logic should ideally just be moved out of the pipeline - extra_step_kwargs = self.prepare_extra_step_kwargs(generator, eta) - - # 7.1 Add image embeds for IP-Adapter - added_cond_kwargs = ( - {"image_embeds": image_embeds} - if ip_adapter_image is not None or ip_adapter_image_embeds is not None - else None - ) - - # 7.2 Create tensor stating which brushnets to keep - brushnet_keep = [] - for i in range(len(timesteps)): - keeps = [ - 1.0 - float(i / len(timesteps) < s or (i + 1) / len(timesteps) > e) - for s, e in zip(control_guidance_start, control_guidance_end) - ] - brushnet_keep.append(keeps[0] if isinstance(brushnet, BrushNetModel) else keeps) - - # 8. Denoising loop - num_warmup_steps = len(timesteps) - num_inference_steps * self.scheduler.order - is_unet_compiled = is_compiled_module(self.unet) - is_brushnet_compiled = is_compiled_module(self.brushnet) - is_torch_higher_equal_2_1 = is_torch_version(">=", "2.1") - with self.progress_bar(total=num_inference_steps) as progress_bar: - for i, t in enumerate(timesteps): - # Relevant thread: - # https://dev-discuss.pytorch.org/t/cudagraphs-in-pytorch-2-0/1428 - if (is_unet_compiled and is_brushnet_compiled) and is_torch_higher_equal_2_1: - torch._inductor.cudagraph_mark_step_begin() - # expand the latents if we are doing classifier free guidance - latent_model_input = torch.cat([latents] * 2) if self.do_classifier_free_guidance else latents - latent_model_input = self.scheduler.scale_model_input(latent_model_input, t) - - # brushnet(s) inference - if guess_mode and self.do_classifier_free_guidance: - # Infer BrushNet only for the conditional batch. - control_model_input = latents - control_model_input = self.scheduler.scale_model_input(control_model_input, t) - brushnet_prompt_embeds = con_prompt_embeds.chunk(2)[1] - else: - control_model_input = latent_model_input - brushnet_prompt_embeds = con_prompt_embeds - - if isinstance(brushnet_keep[i], list): - cond_scale = [c * s for c, s in zip(brushnet_conditioning_scale, brushnet_keep[i])] - else: - brushnet_cond_scale = brushnet_conditioning_scale - if isinstance(brushnet_cond_scale, list): - brushnet_cond_scale = brushnet_cond_scale[0] - cond_scale = brushnet_cond_scale * brushnet_keep[i] - - down_block_res_samples, mid_block_res_sample, up_block_res_samples = self.brushnet( - control_model_input, - t, - encoder_hidden_states=brushnet_prompt_embeds, - brushnet_cond=conditioning_latents, - conditioning_scale=cond_scale, - guess_mode=guess_mode, - return_dict=False, - ) - - if guess_mode and self.do_classifier_free_guidance: - # Infered BrushNet only for the conditional batch. - # To apply the output of BrushNet to both the unconditional and conditional batches, - # add 0 to the unconditional batch to keep it unchanged. - down_block_res_samples = [torch.cat([torch.zeros_like(d), d]) for d in down_block_res_samples] - mid_block_res_sample = torch.cat([torch.zeros_like(mid_block_res_sample), mid_block_res_sample]) - up_block_res_samples = [torch.cat([torch.zeros_like(d), d]) for d in up_block_res_samples] - - # predict the noise residual - noise_pred = self.unet( - latent_model_input, - t, - encoder_hidden_states=prompt_embeds, - timestep_cond=timestep_cond, - cross_attention_kwargs=self.cross_attention_kwargs, - down_block_add_samples=down_block_res_samples, - mid_block_add_sample=mid_block_res_sample, - up_block_add_samples=up_block_res_samples, - added_cond_kwargs=added_cond_kwargs, - return_dict=False, - )[0] - - # perform guidance - if self.do_classifier_free_guidance: - noise_pred_uncond, noise_pred_text = noise_pred.chunk(2) - noise_pred = noise_pred_uncond + self.guidance_scale * (noise_pred_text - noise_pred_uncond) - - # compute the previous noisy sample x_t -> x_t-1 - latents = self.scheduler.step(noise_pred, t, latents, **extra_step_kwargs, return_dict=False)[0] - - if callback_on_step_end is not None: - callback_kwargs = {} - for k in callback_on_step_end_tensor_inputs: - callback_kwargs[k] = locals()[k] - callback_outputs = callback_on_step_end(self, i, t, callback_kwargs) - - latents = callback_outputs.pop("latents", latents) - prompt_embeds = callback_outputs.pop("prompt_embeds", prompt_embeds) - negative_prompt_embeds = callback_outputs.pop("negative_prompt_embeds", negative_prompt_embeds) - - # call the callback, if provided - if i == len(timesteps) - 1 or ((i + 1) > num_warmup_steps and (i + 1) % self.scheduler.order == 0): - progress_bar.update() - if callback is not None and i % callback_steps == 0: - step_idx = i // getattr(self.scheduler, "order", 1) - callback(step_idx, t, latents) - - # If we do sequential model offloading, let's offload unet and brushnet - # manually for max memory savings - if hasattr(self, "final_offload_hook") and self.final_offload_hook is not None: - self.unet.to("cpu") - self.brushnet.to("cpu") - torch.cuda.empty_cache() - - if not output_type == "latent": - image = self.vae.decode(latents / self.vae.config.scaling_factor, return_dict=False, generator=generator)[ - 0 - ] - image, has_nsfw_concept = self.run_safety_checker(image, device, prompt_embeds.dtype) - else: - image = latents - has_nsfw_concept = None - - if has_nsfw_concept is None: - do_denormalize = [True] * image.shape[0] - else: - do_denormalize = [not has_nsfw for has_nsfw in has_nsfw_concept] - - image = self.image_processor.postprocess(image, output_type=output_type, do_denormalize=do_denormalize) - - # Offload all models - self.maybe_free_model_hooks() - - if not return_dict: - return (image, has_nsfw_concept) - - return StableDiffusionPipelineOutput(images=image, nsfw_content_detected=has_nsfw_concept) diff --git a/brushnet/pipeline_brushnet_sd_xl.py b/brushnet/pipeline_brushnet_sd_xl.py deleted file mode 100644 index 8608e2e..0000000 --- a/brushnet/pipeline_brushnet_sd_xl.py +++ /dev/null @@ -1,1539 +0,0 @@ -# Copyright 2024 The HuggingFace Team. All rights reserved. -# -import inspect -from typing import Any, Callable, Dict, List, Optional, Tuple, Union - -import numpy as np -import PIL.Image -import torch -import torch.nn.functional as F - -from transformers import ( - CLIPImageProcessor, - CLIPTextModel, - CLIPTextModelWithProjection, - CLIPTokenizer, - CLIPVisionModelWithProjection, -) - -from diffusers.utils.import_utils import is_invisible_watermark_available - -from diffusers.image_processor import PipelineImageInput, VaeImageProcessor -from diffusers.loaders import ( - FromSingleFileMixin, - IPAdapterMixin, - StableDiffusionXLLoraLoaderMixin, - TextualInversionLoaderMixin, -) -from diffusers.models import AutoencoderKL, ImageProjection -from .brushnet import BrushNetModel -from .unet_2d_condition import UNet2DConditionModel -from diffusers.models.attention_processor import ( - AttnProcessor2_0, - LoRAAttnProcessor2_0, - LoRAXFormersAttnProcessor, - XFormersAttnProcessor, -) -from diffusers.models.lora import adjust_lora_scale_text_encoder -from diffusers.schedulers import KarrasDiffusionSchedulers -from diffusers.utils import ( - USE_PEFT_BACKEND, - deprecate, - logging, - replace_example_docstring, - scale_lora_layers, - unscale_lora_layers, -) -from diffusers.utils.torch_utils import is_compiled_module, is_torch_version, randn_tensor -from diffusers.pipelines.pipeline_utils import DiffusionPipeline, StableDiffusionMixin -from diffusers.pipelines.stable_diffusion_xl.pipeline_output import StableDiffusionXLPipelineOutput - - -if is_invisible_watermark_available(): - from diffusers.pipelines.stable_diffusion_xl.watermark import StableDiffusionXLWatermarker - - -# from .multibrushnet import MultiBrushNetModel - -logger = logging.get_logger(__name__) # pylint: disable=invalid-name - - -EXAMPLE_DOC_STRING = """ - Examples: - ```py - >>> # !pip install opencv-python transformers accelerate - >>> from diffusers import StableDiffusionXLBrushNetPipeline, BrushNetModel, AutoencoderKL - >>> from diffusers.utils import load_image - >>> import numpy as np - >>> import torch - - >>> import cv2 - >>> from PIL import Image - - >>> prompt = "aerial view, a futuristic research complex in a bright foggy jungle, hard lighting" - >>> negative_prompt = "low quality, bad quality, sketches" - - >>> # download an image - >>> image = load_image( - ... "https://hf.co/datasets/hf-internal-testing/diffusers-images/resolve/main/sd_brushnet/hf-logo.png" - ... ) - - >>> # initialize the models and pipeline - >>> brushnet_conditioning_scale = 0.5 # recommended for good generalization - >>> brushnet = BrushNetModel.from_pretrained( - ... "diffusers/brushnet-canny-sdxl-1.0", torch_dtype=torch.float16 - ... ) - >>> vae = AutoencoderKL.from_pretrained("madebyollin/sdxl-vae-fp16-fix", torch_dtype=torch.float16) - >>> pipe = StableDiffusionXLBrushNetPipeline.from_pretrained( - ... "stabilityai/stable-diffusion-xl-base-1.0", brushnet=brushnet, vae=vae, torch_dtype=torch.float16 - ... ) - >>> pipe.enable_model_cpu_offload() - - >>> # get canny image - >>> image = np.array(image) - >>> image = cv2.Canny(image, 100, 200) - >>> image = image[:, :, None] - >>> image = np.concatenate([image, image, image], axis=2) - >>> canny_image = Image.fromarray(image) - - >>> # generate image - >>> image = pipe( - ... prompt, brushnet_conditioning_scale=brushnet_conditioning_scale, image=canny_image - ... ).images[0] - ``` -""" - - -class StableDiffusionXLBrushNetPipeline( - DiffusionPipeline, - StableDiffusionMixin, - TextualInversionLoaderMixin, - StableDiffusionXLLoraLoaderMixin, - IPAdapterMixin, - FromSingleFileMixin, -): - r""" - Pipeline for text-to-image generation using Stable Diffusion XL with BrushNet guidance. - - This model inherits from [`DiffusionPipeline`]. Check the superclass documentation for the generic methods - implemented for all pipelines (downloading, saving, running on a particular device, etc.). - - The pipeline also inherits the following loading methods: - - [`~loaders.TextualInversionLoaderMixin.load_textual_inversion`] for loading textual inversion embeddings - - [`~loaders.StableDiffusionXLLoraLoaderMixin.load_lora_weights`] for loading LoRA weights - - [`~loaders.StableDiffusionXLLoraLoaderMixin.save_lora_weights`] for saving LoRA weights - - [`~loaders.FromSingleFileMixin.from_single_file`] for loading `.ckpt` files - - [`~loaders.IPAdapterMixin.load_ip_adapter`] for loading IP Adapters - - Args: - vae ([`AutoencoderKL`]): - Variational Auto-Encoder (VAE) model to encode and decode images to and from latent representations. - text_encoder ([`~transformers.CLIPTextModel`]): - Frozen text-encoder ([clip-vit-large-patch14](https://huggingface.co/openai/clip-vit-large-patch14)). - text_encoder_2 ([`~transformers.CLIPTextModelWithProjection`]): - Second frozen text-encoder - ([laion/CLIP-ViT-bigG-14-laion2B-39B-b160k](https://huggingface.co/laion/CLIP-ViT-bigG-14-laion2B-39B-b160k)). - tokenizer ([`~transformers.CLIPTokenizer`]): - A `CLIPTokenizer` to tokenize text. - tokenizer_2 ([`~transformers.CLIPTokenizer`]): - A `CLIPTokenizer` to tokenize text. - unet ([`UNet2DConditionModel`]): - A `UNet2DConditionModel` to denoise the encoded image latents. - brushnet ([`BrushNetModel`] or `List[BrushNetModel]`): - Provides additional conditioning to the `unet` during the denoising process. If you set multiple - BrushNets as a list, the outputs from each BrushNet are added together to create one combined - additional conditioning. - scheduler ([`SchedulerMixin`]): - A scheduler to be used in combination with `unet` to denoise the encoded image latents. Can be one of - [`DDIMScheduler`], [`LMSDiscreteScheduler`], or [`PNDMScheduler`]. - force_zeros_for_empty_prompt (`bool`, *optional*, defaults to `"True"`): - Whether the negative prompt embeddings should always be set to 0. Also see the config of - `stabilityai/stable-diffusion-xl-base-1-0`. - add_watermarker (`bool`, *optional*): - Whether to use the [invisible_watermark](https://github.com/ShieldMnt/invisible-watermark/) library to - watermark output images. If not defined, it defaults to `True` if the package is installed; otherwise no - watermarker is used. - """ - - # leave brushnet out on purpose because it iterates with unet - model_cpu_offload_seq = "text_encoder->text_encoder_2->image_encoder->unet->vae" - _optional_components = [ - "tokenizer", - "tokenizer_2", - "text_encoder", - "text_encoder_2", - "feature_extractor", - "image_encoder", - ] - _callback_tensor_inputs = ["latents", "prompt_embeds", "negative_prompt_embeds"] - - def __init__( - self, - vae: AutoencoderKL, - text_encoder: CLIPTextModel, - text_encoder_2: CLIPTextModelWithProjection, - tokenizer: CLIPTokenizer, - tokenizer_2: CLIPTokenizer, - unet: UNet2DConditionModel, - brushnet: Union[BrushNetModel, List[BrushNetModel], Tuple[BrushNetModel]], #MultiBrushNetModel], - scheduler: KarrasDiffusionSchedulers, - force_zeros_for_empty_prompt: bool = True, - add_watermarker: Optional[bool] = None, - feature_extractor: CLIPImageProcessor = None, - image_encoder: CLIPVisionModelWithProjection = None, - ): - super().__init__() - - # if isinstance(brushnet, (list, tuple)): - # brushnet = MultiBrushNetModel(brushnet) - - self.register_modules( - vae=vae, - text_encoder=text_encoder, - text_encoder_2=text_encoder_2, - tokenizer=tokenizer, - tokenizer_2=tokenizer_2, - unet=unet, - brushnet=brushnet, - scheduler=scheduler, - feature_extractor=feature_extractor, - image_encoder=image_encoder, - ) - self.vae_scale_factor = 2 ** (len(self.vae.config.block_out_channels) - 1) - self.image_processor = VaeImageProcessor(vae_scale_factor=self.vae_scale_factor, do_convert_rgb=True) - # self.control_image_processor = VaeImageProcessor( - # vae_scale_factor=self.vae_scale_factor, do_convert_rgb=True, do_normalize=False - # ) - add_watermarker = add_watermarker if add_watermarker is not None else is_invisible_watermark_available() - - if add_watermarker: - self.watermark = StableDiffusionXLWatermarker() - else: - self.watermark = None - - self.register_to_config(force_zeros_for_empty_prompt=force_zeros_for_empty_prompt) - - # Copied from diffusers.pipelines.stable_diffusion_xl.pipeline_stable_diffusion_xl.StableDiffusionXLPipeline.encode_prompt - def encode_prompt( - self, - prompt: str, - prompt_2: Optional[str] = None, - device: Optional[torch.device] = None, - num_images_per_prompt: int = 1, - do_classifier_free_guidance: bool = True, - negative_prompt: Optional[str] = None, - negative_prompt_2: Optional[str] = None, - prompt_embeds: Optional[torch.FloatTensor] = None, - negative_prompt_embeds: Optional[torch.FloatTensor] = None, - pooled_prompt_embeds: Optional[torch.FloatTensor] = None, - negative_pooled_prompt_embeds: Optional[torch.FloatTensor] = None, - lora_scale: Optional[float] = None, - clip_skip: Optional[int] = None, - ): - r""" - Encodes the prompt into text encoder hidden states. - - Args: - prompt (`str` or `List[str]`, *optional*): - prompt to be encoded - prompt_2 (`str` or `List[str]`, *optional*): - The prompt or prompts to be sent to the `tokenizer_2` and `text_encoder_2`. If not defined, `prompt` is - used in both text-encoders - device: (`torch.device`): - torch device - num_images_per_prompt (`int`): - number of images that should be generated per prompt - do_classifier_free_guidance (`bool`): - whether to use classifier free guidance or not - negative_prompt (`str` or `List[str]`, *optional*): - The prompt or prompts not to guide the image generation. If not defined, one has to pass - `negative_prompt_embeds` instead. Ignored when not using guidance (i.e., ignored if `guidance_scale` is - less than `1`). - negative_prompt_2 (`str` or `List[str]`, *optional*): - The prompt or prompts not to guide the image generation to be sent to `tokenizer_2` and - `text_encoder_2`. If not defined, `negative_prompt` is used in both text-encoders - prompt_embeds (`torch.FloatTensor`, *optional*): - Pre-generated text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt weighting. If not - provided, text embeddings will be generated from `prompt` input argument. - negative_prompt_embeds (`torch.FloatTensor`, *optional*): - Pre-generated negative text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt - weighting. If not provided, negative_prompt_embeds will be generated from `negative_prompt` input - argument. - pooled_prompt_embeds (`torch.FloatTensor`, *optional*): - Pre-generated pooled text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt weighting. - If not provided, pooled text embeddings will be generated from `prompt` input argument. - negative_pooled_prompt_embeds (`torch.FloatTensor`, *optional*): - Pre-generated negative pooled text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt - weighting. If not provided, pooled negative_prompt_embeds will be generated from `negative_prompt` - input argument. - lora_scale (`float`, *optional*): - A lora scale that will be applied to all LoRA layers of the text encoder if LoRA layers are loaded. - clip_skip (`int`, *optional*): - Number of layers to be skipped from CLIP while computing the prompt embeddings. A value of 1 means that - the output of the pre-final layer will be used for computing the prompt embeddings. - """ - device = device or self._execution_device - - # set lora scale so that monkey patched LoRA - # function of text encoder can correctly access it - if lora_scale is not None and isinstance(self, StableDiffusionXLLoraLoaderMixin): - self._lora_scale = lora_scale - - # dynamically adjust the LoRA scale - if self.text_encoder is not None: - if not USE_PEFT_BACKEND: - adjust_lora_scale_text_encoder(self.text_encoder, lora_scale) - else: - scale_lora_layers(self.text_encoder, lora_scale) - - if self.text_encoder_2 is not None: - if not USE_PEFT_BACKEND: - adjust_lora_scale_text_encoder(self.text_encoder_2, lora_scale) - else: - scale_lora_layers(self.text_encoder_2, lora_scale) - - prompt = [prompt] if isinstance(prompt, str) else prompt - - if prompt is not None: - batch_size = len(prompt) - else: - batch_size = prompt_embeds.shape[0] - - # Define tokenizers and text encoders - tokenizers = [self.tokenizer, self.tokenizer_2] if self.tokenizer is not None else [self.tokenizer_2] - text_encoders = ( - [self.text_encoder, self.text_encoder_2] if self.text_encoder is not None else [self.text_encoder_2] - ) - - if prompt_embeds is None: - prompt_2 = prompt_2 or prompt - prompt_2 = [prompt_2] if isinstance(prompt_2, str) else prompt_2 - - # textual inversion: process multi-vector tokens if necessary - prompt_embeds_list = [] - prompts = [prompt, prompt_2] - for prompt, tokenizer, text_encoder in zip(prompts, tokenizers, text_encoders): - if isinstance(self, TextualInversionLoaderMixin): - prompt = self.maybe_convert_prompt(prompt, tokenizer) - - text_inputs = tokenizer( - prompt, - padding="max_length", - max_length=tokenizer.model_max_length, - truncation=True, - return_tensors="pt", - ) - - text_input_ids = text_inputs.input_ids - untruncated_ids = tokenizer(prompt, padding="longest", return_tensors="pt").input_ids - - if untruncated_ids.shape[-1] >= text_input_ids.shape[-1] and not torch.equal( - text_input_ids, untruncated_ids - ): - removed_text = tokenizer.batch_decode(untruncated_ids[:, tokenizer.model_max_length - 1 : -1]) - logger.warning( - "The following part of your input was truncated because CLIP can only handle sequences up to" - f" {tokenizer.model_max_length} tokens: {removed_text}" - ) - - prompt_embeds = text_encoder(text_input_ids.to(device), output_hidden_states=True) - - # We are only ALWAYS interested in the pooled output of the final text encoder - pooled_prompt_embeds = prompt_embeds[0] - if clip_skip is None: - prompt_embeds = prompt_embeds.hidden_states[-2] - else: - # "2" because SDXL always indexes from the penultimate layer. - prompt_embeds = prompt_embeds.hidden_states[-(clip_skip + 2)] - - prompt_embeds_list.append(prompt_embeds) - - prompt_embeds = torch.concat(prompt_embeds_list, dim=-1) - - # get unconditional embeddings for classifier free guidance - zero_out_negative_prompt = negative_prompt is None and self.config.force_zeros_for_empty_prompt - if do_classifier_free_guidance and negative_prompt_embeds is None and zero_out_negative_prompt: - negative_prompt_embeds = torch.zeros_like(prompt_embeds) - negative_pooled_prompt_embeds = torch.zeros_like(pooled_prompt_embeds) - elif do_classifier_free_guidance and negative_prompt_embeds is None: - negative_prompt = negative_prompt or "" - negative_prompt_2 = negative_prompt_2 or negative_prompt - - # normalize str to list - negative_prompt = batch_size * [negative_prompt] if isinstance(negative_prompt, str) else negative_prompt - negative_prompt_2 = ( - batch_size * [negative_prompt_2] if isinstance(negative_prompt_2, str) else negative_prompt_2 - ) - - uncond_tokens: List[str] - if prompt is not None and type(prompt) is not type(negative_prompt): - raise TypeError( - f"`negative_prompt` should be the same type to `prompt`, but got {type(negative_prompt)} !=" - f" {type(prompt)}." - ) - elif batch_size != len(negative_prompt): - raise ValueError( - f"`negative_prompt`: {negative_prompt} has batch size {len(negative_prompt)}, but `prompt`:" - f" {prompt} has batch size {batch_size}. Please make sure that passed `negative_prompt` matches" - " the batch size of `prompt`." - ) - else: - uncond_tokens = [negative_prompt, negative_prompt_2] - - negative_prompt_embeds_list = [] - for negative_prompt, tokenizer, text_encoder in zip(uncond_tokens, tokenizers, text_encoders): - if isinstance(self, TextualInversionLoaderMixin): - negative_prompt = self.maybe_convert_prompt(negative_prompt, tokenizer) - - max_length = prompt_embeds.shape[1] - uncond_input = tokenizer( - negative_prompt, - padding="max_length", - max_length=max_length, - truncation=True, - return_tensors="pt", - ) - - negative_prompt_embeds = text_encoder( - uncond_input.input_ids.to(device), - output_hidden_states=True, - ) - # We are only ALWAYS interested in the pooled output of the final text encoder - negative_pooled_prompt_embeds = negative_prompt_embeds[0] - negative_prompt_embeds = negative_prompt_embeds.hidden_states[-2] - - negative_prompt_embeds_list.append(negative_prompt_embeds) - - negative_prompt_embeds = torch.concat(negative_prompt_embeds_list, dim=-1) - - if self.text_encoder_2 is not None: - prompt_embeds = prompt_embeds.to(dtype=self.text_encoder_2.dtype, device=device) - else: - prompt_embeds = prompt_embeds.to(dtype=self.unet.dtype, device=device) - - bs_embed, seq_len, _ = prompt_embeds.shape - # duplicate text embeddings for each generation per prompt, using mps friendly method - prompt_embeds = prompt_embeds.repeat(1, num_images_per_prompt, 1) - prompt_embeds = prompt_embeds.view(bs_embed * num_images_per_prompt, seq_len, -1) - - if do_classifier_free_guidance: - # duplicate unconditional embeddings for each generation per prompt, using mps friendly method - seq_len = negative_prompt_embeds.shape[1] - - if self.text_encoder_2 is not None: - negative_prompt_embeds = negative_prompt_embeds.to(dtype=self.text_encoder_2.dtype, device=device) - else: - negative_prompt_embeds = negative_prompt_embeds.to(dtype=self.unet.dtype, device=device) - - negative_prompt_embeds = negative_prompt_embeds.repeat(1, num_images_per_prompt, 1) - negative_prompt_embeds = negative_prompt_embeds.view(batch_size * num_images_per_prompt, seq_len, -1) - - pooled_prompt_embeds = pooled_prompt_embeds.repeat(1, num_images_per_prompt).view( - bs_embed * num_images_per_prompt, -1 - ) - if do_classifier_free_guidance: - negative_pooled_prompt_embeds = negative_pooled_prompt_embeds.repeat(1, num_images_per_prompt).view( - bs_embed * num_images_per_prompt, -1 - ) - - if self.text_encoder is not None: - if isinstance(self, StableDiffusionXLLoraLoaderMixin) and USE_PEFT_BACKEND: - # Retrieve the original scale by scaling back the LoRA layers - unscale_lora_layers(self.text_encoder, lora_scale) - - if self.text_encoder_2 is not None: - if isinstance(self, StableDiffusionXLLoraLoaderMixin) and USE_PEFT_BACKEND: - # Retrieve the original scale by scaling back the LoRA layers - unscale_lora_layers(self.text_encoder_2, lora_scale) - - return prompt_embeds, negative_prompt_embeds, pooled_prompt_embeds, negative_pooled_prompt_embeds - - # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.encode_image - def encode_image(self, image, device, num_images_per_prompt, output_hidden_states=None): - dtype = next(self.image_encoder.parameters()).dtype - - if not isinstance(image, torch.Tensor): - image = self.feature_extractor(image, return_tensors="pt").pixel_values - - image = image.to(device=device, dtype=dtype) - if output_hidden_states: - image_enc_hidden_states = self.image_encoder(image, output_hidden_states=True).hidden_states[-2] - image_enc_hidden_states = image_enc_hidden_states.repeat_interleave(num_images_per_prompt, dim=0) - uncond_image_enc_hidden_states = self.image_encoder( - torch.zeros_like(image), output_hidden_states=True - ).hidden_states[-2] - uncond_image_enc_hidden_states = uncond_image_enc_hidden_states.repeat_interleave( - num_images_per_prompt, dim=0 - ) - return image_enc_hidden_states, uncond_image_enc_hidden_states - else: - image_embeds = self.image_encoder(image).image_embeds - image_embeds = image_embeds.repeat_interleave(num_images_per_prompt, dim=0) - uncond_image_embeds = torch.zeros_like(image_embeds) - - return image_embeds, uncond_image_embeds - - # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.prepare_ip_adapter_image_embeds - def prepare_ip_adapter_image_embeds( - self, ip_adapter_image, ip_adapter_image_embeds, device, num_images_per_prompt, do_classifier_free_guidance - ): - if ip_adapter_image_embeds is None: - if not isinstance(ip_adapter_image, list): - ip_adapter_image = [ip_adapter_image] - - if len(ip_adapter_image) != len(self.unet.encoder_hid_proj.image_projection_layers): - raise ValueError( - f"`ip_adapter_image` must have same length as the number of IP Adapters. Got {len(ip_adapter_image)} images and {len(self.unet.encoder_hid_proj.image_projection_layers)} IP Adapters." - ) - - image_embeds = [] - for single_ip_adapter_image, image_proj_layer in zip( - ip_adapter_image, self.unet.encoder_hid_proj.image_projection_layers - ): - output_hidden_state = not isinstance(image_proj_layer, ImageProjection) - single_image_embeds, single_negative_image_embeds = self.encode_image( - single_ip_adapter_image, device, 1, output_hidden_state - ) - single_image_embeds = torch.stack([single_image_embeds] * num_images_per_prompt, dim=0) - single_negative_image_embeds = torch.stack( - [single_negative_image_embeds] * num_images_per_prompt, dim=0 - ) - - if do_classifier_free_guidance: - single_image_embeds = torch.cat([single_negative_image_embeds, single_image_embeds]) - single_image_embeds = single_image_embeds.to(device) - - image_embeds.append(single_image_embeds) - else: - repeat_dims = [1] - image_embeds = [] - for single_image_embeds in ip_adapter_image_embeds: - if do_classifier_free_guidance: - single_negative_image_embeds, single_image_embeds = single_image_embeds.chunk(2) - single_image_embeds = single_image_embeds.repeat( - num_images_per_prompt, *(repeat_dims * len(single_image_embeds.shape[1:])) - ) - single_negative_image_embeds = single_negative_image_embeds.repeat( - num_images_per_prompt, *(repeat_dims * len(single_negative_image_embeds.shape[1:])) - ) - single_image_embeds = torch.cat([single_negative_image_embeds, single_image_embeds]) - else: - single_image_embeds = single_image_embeds.repeat( - num_images_per_prompt, *(repeat_dims * len(single_image_embeds.shape[1:])) - ) - image_embeds.append(single_image_embeds) - - return image_embeds - - # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.prepare_extra_step_kwargs - def prepare_extra_step_kwargs(self, generator, eta): - # prepare extra kwargs for the scheduler step, since not all schedulers have the same signature - # eta (η) is only used with the DDIMScheduler, it will be ignored for other schedulers. - # eta corresponds to η in DDIM paper: https://arxiv.org/abs/2010.02502 - # and should be between [0, 1] - - accepts_eta = "eta" in set(inspect.signature(self.scheduler.step).parameters.keys()) - extra_step_kwargs = {} - if accepts_eta: - extra_step_kwargs["eta"] = eta - - # check if the scheduler accepts generator - accepts_generator = "generator" in set(inspect.signature(self.scheduler.step).parameters.keys()) - if accepts_generator: - extra_step_kwargs["generator"] = generator - return extra_step_kwargs - - def check_inputs( - self, - prompt, - prompt_2, - image, - mask, - callback_steps, - negative_prompt=None, - negative_prompt_2=None, - prompt_embeds=None, - negative_prompt_embeds=None, - pooled_prompt_embeds=None, - ip_adapter_image=None, - ip_adapter_image_embeds=None, - negative_pooled_prompt_embeds=None, - brushnet_conditioning_scale=1.0, - control_guidance_start=0.0, - control_guidance_end=1.0, - callback_on_step_end_tensor_inputs=None, - ): - if callback_steps is not None and (not isinstance(callback_steps, int) or callback_steps <= 0): - raise ValueError( - f"`callback_steps` has to be a positive integer but is {callback_steps} of type" - f" {type(callback_steps)}." - ) - - if callback_on_step_end_tensor_inputs is not None and not all( - k in self._callback_tensor_inputs for k in callback_on_step_end_tensor_inputs - ): - raise ValueError( - f"`callback_on_step_end_tensor_inputs` has to be in {self._callback_tensor_inputs}, but found {[k for k in callback_on_step_end_tensor_inputs if k not in self._callback_tensor_inputs]}" - ) - - if prompt is not None and prompt_embeds is not None: - raise ValueError( - f"Cannot forward both `prompt`: {prompt} and `prompt_embeds`: {prompt_embeds}. Please make sure to" - " only forward one of the two." - ) - elif prompt_2 is not None and prompt_embeds is not None: - raise ValueError( - f"Cannot forward both `prompt_2`: {prompt_2} and `prompt_embeds`: {prompt_embeds}. Please make sure to" - " only forward one of the two." - ) - elif prompt is None and prompt_embeds is None: - raise ValueError( - "Provide either `prompt` or `prompt_embeds`. Cannot leave both `prompt` and `prompt_embeds` undefined." - ) - elif prompt is not None and (not isinstance(prompt, str) and not isinstance(prompt, list)): - raise ValueError(f"`prompt` has to be of type `str` or `list` but is {type(prompt)}") - elif prompt_2 is not None and (not isinstance(prompt_2, str) and not isinstance(prompt_2, list)): - raise ValueError(f"`prompt_2` has to be of type `str` or `list` but is {type(prompt_2)}") - - if negative_prompt is not None and negative_prompt_embeds is not None: - raise ValueError( - f"Cannot forward both `negative_prompt`: {negative_prompt} and `negative_prompt_embeds`:" - f" {negative_prompt_embeds}. Please make sure to only forward one of the two." - ) - elif negative_prompt_2 is not None and negative_prompt_embeds is not None: - raise ValueError( - f"Cannot forward both `negative_prompt_2`: {negative_prompt_2} and `negative_prompt_embeds`:" - f" {negative_prompt_embeds}. Please make sure to only forward one of the two." - ) - - if prompt_embeds is not None and negative_prompt_embeds is not None: - if prompt_embeds.shape != negative_prompt_embeds.shape: - raise ValueError( - "`prompt_embeds` and `negative_prompt_embeds` must have the same shape when passed directly, but" - f" got: `prompt_embeds` {prompt_embeds.shape} != `negative_prompt_embeds`" - f" {negative_prompt_embeds.shape}." - ) - - if prompt_embeds is not None and pooled_prompt_embeds is None: - raise ValueError( - "If `prompt_embeds` are provided, `pooled_prompt_embeds` also have to be passed. Make sure to generate `pooled_prompt_embeds` from the same text encoder that was used to generate `prompt_embeds`." - ) - - if negative_prompt_embeds is not None and negative_pooled_prompt_embeds is None: - raise ValueError( - "If `negative_prompt_embeds` are provided, `negative_pooled_prompt_embeds` also have to be passed. Make sure to generate `negative_pooled_prompt_embeds` from the same text encoder that was used to generate `negative_prompt_embeds`." - ) - - # `prompt` needs more sophisticated handling when there are multiple - # conditionings. - # if isinstance(self.brushnet, MultiBrushNetModel): - # if isinstance(prompt, list): - # logger.warning( - # f"You have {len(self.brushnet.nets)} BrushNets and you have passed {len(prompt)}" - # " prompts. The conditionings will be fixed across the prompts." - # ) - - # Check `image` - is_compiled = hasattr(F, "scaled_dot_product_attention") and isinstance( - self.brushnet, torch._dynamo.eval_frame.OptimizedModule - ) - if ( - isinstance(self.brushnet, BrushNetModel) - or is_compiled - and isinstance(self.brushnet._orig_mod, BrushNetModel) - ): - self.check_image(image, mask, prompt, prompt_embeds) - # elif ( - # isinstance(self.brushnet, MultiBrushNetModel) - # or is_compiled - # and isinstance(self.brushnet._orig_mod, MultiBrushNetModel) - # ): - # if not isinstance(image, list): - # raise TypeError("For multiple brushnets: `image` must be type `list`") - - # # When `image` is a nested list: - # # (e.g. [[canny_image_1, pose_image_1], [canny_image_2, pose_image_2]]) - # elif any(isinstance(i, list) for i in image): - # raise ValueError("A single batch of multiple conditionings are supported at the moment.") - # elif len(image) != len(self.brushnet.nets): - # raise ValueError( - # f"For multiple brushnets: `image` must have the same length as the number of brushnets, but got {len(image)} images and {len(self.brushnet.nets)} BrushNets." - # ) - - # for image_ in image: - # self.check_image(image_, prompt, prompt_embeds) - else: - assert False - - # Check `brushnet_conditioning_scale` - if ( - isinstance(self.brushnet, BrushNetModel) - or is_compiled - and isinstance(self.brushnet._orig_mod, BrushNetModel) - ): - if not isinstance(brushnet_conditioning_scale, float): - raise TypeError("For single brushnet: `brushnet_conditioning_scale` must be type `float`.") - # elif ( - # isinstance(self.brushnet, MultiBrushNetModel) - # or is_compiled - # and isinstance(self.brushnet._orig_mod, MultiBrushNetModel) - # ): - # if isinstance(brushnet_conditioning_scale, list): - # if any(isinstance(i, list) for i in brushnet_conditioning_scale): - # raise ValueError("A single batch of multiple conditionings are supported at the moment.") - # elif isinstance(brushnet_conditioning_scale, list) and len(brushnet_conditioning_scale) != len( - # self.brushnet.nets - # ): - # raise ValueError( - # "For multiple brushnets: When `brushnet_conditioning_scale` is specified as `list`, it must have" - # " the same length as the number of brushnets" - # ) - else: - assert False - - if not isinstance(control_guidance_start, (tuple, list)): - control_guidance_start = [control_guidance_start] - - if not isinstance(control_guidance_end, (tuple, list)): - control_guidance_end = [control_guidance_end] - - if len(control_guidance_start) != len(control_guidance_end): - raise ValueError( - f"`control_guidance_start` has {len(control_guidance_start)} elements, but `control_guidance_end` has {len(control_guidance_end)} elements. Make sure to provide the same number of elements to each list." - ) - - # if isinstance(self.brushnet, MultiBrushNetModel): - # if len(control_guidance_start) != len(self.brushnet.nets): - # raise ValueError( - # f"`control_guidance_start`: {control_guidance_start} has {len(control_guidance_start)} elements but there are {len(self.brushnet.nets)} brushnets available. Make sure to provide {len(self.brushnet.nets)}." - # ) - - for start, end in zip(control_guidance_start, control_guidance_end): - if start >= end: - raise ValueError( - f"control guidance start: {start} cannot be larger or equal to control guidance end: {end}." - ) - if start < 0.0: - raise ValueError(f"control guidance start: {start} can't be smaller than 0.") - if end > 1.0: - raise ValueError(f"control guidance end: {end} can't be larger than 1.0.") - - if ip_adapter_image is not None and ip_adapter_image_embeds is not None: - raise ValueError( - "Provide either `ip_adapter_image` or `ip_adapter_image_embeds`. Cannot leave both `ip_adapter_image` and `ip_adapter_image_embeds` defined." - ) - - if ip_adapter_image_embeds is not None: - if not isinstance(ip_adapter_image_embeds, list): - raise ValueError( - f"`ip_adapter_image_embeds` has to be of type `list` but is {type(ip_adapter_image_embeds)}" - ) - elif ip_adapter_image_embeds[0].ndim not in [3, 4]: - raise ValueError( - f"`ip_adapter_image_embeds` has to be a list of 3D or 4D tensors but is {ip_adapter_image_embeds[0].ndim}D" - ) - - # Copied from diffusers.pipelines.brushnet.pipeline_brushnet.StableDiffusionBrushNetPipeline.check_image - def check_image(self, image, mask, prompt, prompt_embeds): - image_is_pil = isinstance(image, PIL.Image.Image) - image_is_tensor = isinstance(image, torch.Tensor) - image_is_np = isinstance(image, np.ndarray) - image_is_pil_list = isinstance(image, list) and isinstance(image[0], PIL.Image.Image) - image_is_tensor_list = isinstance(image, list) and isinstance(image[0], torch.Tensor) - image_is_np_list = isinstance(image, list) and isinstance(image[0], np.ndarray) - - if ( - not image_is_pil - and not image_is_tensor - and not image_is_np - and not image_is_pil_list - and not image_is_tensor_list - and not image_is_np_list - ): - raise TypeError( - f"image must be passed and be one of PIL image, numpy array, torch tensor, list of PIL images, list of numpy arrays or list of torch tensors, but is {type(image)}" - ) - - mask_is_pil = isinstance(mask, PIL.Image.Image) - mask_is_tensor = isinstance(mask, torch.Tensor) - mask_is_np = isinstance(mask, np.ndarray) - mask_is_pil_list = isinstance(mask, list) and isinstance(mask[0], PIL.Image.Image) - mask_is_tensor_list = isinstance(mask, list) and isinstance(mask[0], torch.Tensor) - mask_is_np_list = isinstance(mask, list) and isinstance(mask[0], np.ndarray) - - if ( - not mask_is_pil - and not mask_is_tensor - and not mask_is_np - and not mask_is_pil_list - and not mask_is_tensor_list - and not mask_is_np_list - ): - raise TypeError( - f"mask must be passed and be one of PIL image, numpy array, torch tensor, list of PIL images, list of numpy arrays or list of torch tensors, but is {type(mask)}" - ) - - - if image_is_pil: - image_batch_size = 1 - else: - image_batch_size = len(image) - - if prompt is not None and isinstance(prompt, str): - prompt_batch_size = 1 - elif prompt is not None and isinstance(prompt, list): - prompt_batch_size = len(prompt) - elif prompt_embeds is not None: - prompt_batch_size = prompt_embeds.shape[0] - - if image_batch_size != 1 and image_batch_size != prompt_batch_size: - raise ValueError( - f"If image batch size is not 1, image batch size must be same as prompt batch size. image batch size: {image_batch_size}, prompt batch size: {prompt_batch_size}" - ) - - # Copied from diffusers.pipelines.brushnet.pipeline_brushnet.StableDiffusionBrushNetPipeline.prepare_image - def prepare_image( - self, - image, - width, - height, - batch_size, - num_images_per_prompt, - device, - dtype, - do_classifier_free_guidance=False, - guess_mode=False, - ): - image = self.image_processor.preprocess(image, height=height, width=width).to(dtype=torch.float32) - image_batch_size = image.shape[0] - - if image_batch_size == 1: - repeat_by = batch_size - else: - # image batch size is the same as prompt batch size - repeat_by = num_images_per_prompt - - image = image.repeat_interleave(repeat_by, dim=0) - - image = image.to(device=device, dtype=dtype) - - if do_classifier_free_guidance and not guess_mode: - image = torch.cat([image] * 2) - - return image - - # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.prepare_latents - def prepare_latents(self, batch_size, num_channels_latents, height, width, dtype, device, generator, latents=None): - shape = (batch_size, num_channels_latents, height // self.vae_scale_factor, width // self.vae_scale_factor) - if isinstance(generator, list) and len(generator) != batch_size: - raise ValueError( - f"You have passed a list of generators of length {len(generator)}, but requested an effective batch" - f" size of {batch_size}. Make sure the batch size matches the length of the generators." - ) - - if latents is None: - noise = randn_tensor(shape, generator=generator, device=device, dtype=dtype) - else: - noise = latents.to(device) - - # scale the initial noise by the standard deviation required by the scheduler - latents = noise * self.scheduler.init_noise_sigma - return latents, noise - - # Copied from diffusers.pipelines.stable_diffusion_xl.pipeline_stable_diffusion_xl.StableDiffusionXLPipeline._get_add_time_ids - def _get_add_time_ids( - self, original_size, crops_coords_top_left, target_size, dtype, text_encoder_projection_dim=None - ): - add_time_ids = list(original_size + crops_coords_top_left + target_size) - - passed_add_embed_dim = ( - self.unet.config.addition_time_embed_dim * len(add_time_ids) + text_encoder_projection_dim - ) - expected_add_embed_dim = self.unet.add_embedding.linear_1.in_features - - if expected_add_embed_dim != passed_add_embed_dim: - raise ValueError( - f"Model expects an added time embedding vector of length {expected_add_embed_dim}, but a vector of {passed_add_embed_dim} was created. The model has an incorrect config. Please check `unet.config.time_embedding_type` and `text_encoder_2.config.projection_dim`." - ) - - add_time_ids = torch.tensor([add_time_ids], dtype=dtype) - return add_time_ids - - # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion_upscale.StableDiffusionUpscalePipeline.upcast_vae - def upcast_vae(self): - dtype = self.vae.dtype - self.vae.to(dtype=torch.float32) - use_torch_2_0_or_xformers = isinstance( - self.vae.decoder.mid_block.attentions[0].processor, - ( - AttnProcessor2_0, - XFormersAttnProcessor, - LoRAXFormersAttnProcessor, - LoRAAttnProcessor2_0, - ), - ) - # if xformers or torch_2_0 is used attention block does not need - # to be in float32 which can save lots of memory - if use_torch_2_0_or_xformers: - self.vae.post_quant_conv.to(dtype) - self.vae.decoder.conv_in.to(dtype) - self.vae.decoder.mid_block.to(dtype) - - # Copied from diffusers.pipelines.latent_consistency_models.pipeline_latent_consistency_text2img.LatentConsistencyModelPipeline.get_guidance_scale_embedding - def get_guidance_scale_embedding(self, w, embedding_dim=512, dtype=torch.float32): - """ - See https://github.com/google-research/vdm/blob/dc27b98a554f65cdc654b800da5aa1846545d41b/model_vdm.py#L298 - - Args: - timesteps (`torch.Tensor`): - generate embedding vectors at these timesteps - embedding_dim (`int`, *optional*, defaults to 512): - dimension of the embeddings to generate - dtype: - data type of the generated embeddings - - Returns: - `torch.FloatTensor`: Embedding vectors with shape `(len(timesteps), embedding_dim)` - """ - assert len(w.shape) == 1 - w = w * 1000.0 - - half_dim = embedding_dim // 2 - emb = torch.log(torch.tensor(10000.0)) / (half_dim - 1) - emb = torch.exp(torch.arange(half_dim, dtype=dtype) * -emb) - emb = w.to(dtype)[:, None] * emb[None, :] - emb = torch.cat([torch.sin(emb), torch.cos(emb)], dim=1) - if embedding_dim % 2 == 1: # zero pad - emb = torch.nn.functional.pad(emb, (0, 1)) - assert emb.shape == (w.shape[0], embedding_dim) - return emb - - @property - def guidance_scale(self): - return self._guidance_scale - - @property - def clip_skip(self): - return self._clip_skip - - # here `guidance_scale` is defined analog to the guidance weight `w` of equation (2) - # of the Imagen paper: https://arxiv.org/pdf/2205.11487.pdf . `guidance_scale = 1` - # corresponds to doing no classifier free guidance. - @property - def do_classifier_free_guidance(self): - return self._guidance_scale > 1 and self.unet.config.time_cond_proj_dim is None - - @property - def cross_attention_kwargs(self): - return self._cross_attention_kwargs - - @property - def denoising_end(self): - return self._denoising_end - - @property - def num_timesteps(self): - return self._num_timesteps - - @torch.no_grad() - @replace_example_docstring(EXAMPLE_DOC_STRING) - def __call__( - self, - prompt: Union[str, List[str]] = None, - prompt_2: Optional[Union[str, List[str]]] = None, - image: PipelineImageInput = None, - mask: PipelineImageInput = None, - height: Optional[int] = None, - width: Optional[int] = None, - num_inference_steps: int = 50, - denoising_end: Optional[float] = None, - guidance_scale: float = 5.0, - negative_prompt: Optional[Union[str, List[str]]] = None, - negative_prompt_2: Optional[Union[str, List[str]]] = None, - num_images_per_prompt: Optional[int] = 1, - eta: float = 0.0, - generator: Optional[Union[torch.Generator, List[torch.Generator]]] = None, - latents: Optional[torch.FloatTensor] = None, - prompt_embeds: Optional[torch.FloatTensor] = None, - negative_prompt_embeds: Optional[torch.FloatTensor] = None, - pooled_prompt_embeds: Optional[torch.FloatTensor] = None, - negative_pooled_prompt_embeds: Optional[torch.FloatTensor] = None, - ip_adapter_image: Optional[PipelineImageInput] = None, - ip_adapter_image_embeds: Optional[List[torch.FloatTensor]] = None, - output_type: Optional[str] = "pil", - return_dict: bool = True, - cross_attention_kwargs: Optional[Dict[str, Any]] = None, - brushnet_conditioning_scale: Union[float, List[float]] = 1.0, - guess_mode: bool = False, - control_guidance_start: Union[float, List[float]] = 0.0, - control_guidance_end: Union[float, List[float]] = 1.0, - original_size: Tuple[int, int] = None, - crops_coords_top_left: Tuple[int, int] = (0, 0), - target_size: Tuple[int, int] = None, - negative_original_size: Optional[Tuple[int, int]] = None, - negative_crops_coords_top_left: Tuple[int, int] = (0, 0), - negative_target_size: Optional[Tuple[int, int]] = None, - clip_skip: Optional[int] = None, - callback_on_step_end: Optional[Callable[[int, int, Dict], None]] = None, - callback_on_step_end_tensor_inputs: List[str] = ["latents"], - **kwargs, - ): - r""" - The call function to the pipeline for generation. - - Args: - prompt (`str` or `List[str]`, *optional*): - The prompt or prompts to guide image generation. If not defined, you need to pass `prompt_embeds`. - prompt_2 (`str` or `List[str]`, *optional*): - The prompt or prompts to be sent to `tokenizer_2` and `text_encoder_2`. If not defined, `prompt` is - used in both text-encoders. - image (`torch.FloatTensor`, `PIL.Image.Image`, `np.ndarray`, `List[torch.FloatTensor]`, `List[PIL.Image.Image]`, `List[np.ndarray]`,: - `List[List[torch.FloatTensor]]`, `List[List[np.ndarray]]` or `List[List[PIL.Image.Image]]`): - The BrushNet input condition to provide guidance to the `unet` for generation. If the type is - specified as `torch.FloatTensor`, it is passed to BrushNet as is. `PIL.Image.Image` can also be - accepted as an image. The dimensions of the output image defaults to `image`'s dimensions. If height - and/or width are passed, `image` is resized accordingly. If multiple BrushNets are specified in - `init`, images must be passed as a list such that each element of the list can be correctly batched for - input to a single BrushNet. - height (`int`, *optional*, defaults to `self.unet.config.sample_size * self.vae_scale_factor`): - The height in pixels of the generated image. Anything below 512 pixels won't work well for - [stabilityai/stable-diffusion-xl-base-1.0](https://huggingface.co/stabilityai/stable-diffusion-xl-base-1.0) - and checkpoints that are not specifically fine-tuned on low resolutions. - width (`int`, *optional*, defaults to `self.unet.config.sample_size * self.vae_scale_factor`): - The width in pixels of the generated image. Anything below 512 pixels won't work well for - [stabilityai/stable-diffusion-xl-base-1.0](https://huggingface.co/stabilityai/stable-diffusion-xl-base-1.0) - and checkpoints that are not specifically fine-tuned on low resolutions. - num_inference_steps (`int`, *optional*, defaults to 50): - The number of denoising steps. More denoising steps usually lead to a higher quality image at the - expense of slower inference. - denoising_end (`float`, *optional*): - When specified, determines the fraction (between 0.0 and 1.0) of the total denoising process to be - completed before it is intentionally prematurely terminated. As a result, the returned sample will - still retain a substantial amount of noise as determined by the discrete timesteps selected by the - scheduler. The denoising_end parameter should ideally be utilized when this pipeline forms a part of a - "Mixture of Denoisers" multi-pipeline setup, as elaborated in [**Refining the Image - Output**](https://huggingface.co/docs/diffusers/api/pipelines/stable_diffusion/stable_diffusion_xl#refining-the-image-output) - guidance_scale (`float`, *optional*, defaults to 5.0): - A higher guidance scale value encourages the model to generate images closely linked to the text - `prompt` at the expense of lower image quality. Guidance scale is enabled when `guidance_scale > 1`. - negative_prompt (`str` or `List[str]`, *optional*): - The prompt or prompts to guide what to not include in image generation. If not defined, you need to - pass `negative_prompt_embeds` instead. Ignored when not using guidance (`guidance_scale < 1`). - negative_prompt_2 (`str` or `List[str]`, *optional*): - The prompt or prompts to guide what to not include in image generation. This is sent to `tokenizer_2` - and `text_encoder_2`. If not defined, `negative_prompt` is used in both text-encoders. - num_images_per_prompt (`int`, *optional*, defaults to 1): - The number of images to generate per prompt. - eta (`float`, *optional*, defaults to 0.0): - Corresponds to parameter eta (η) from the [DDIM](https://arxiv.org/abs/2010.02502) paper. Only applies - to the [`~schedulers.DDIMScheduler`], and is ignored in other schedulers. - generator (`torch.Generator` or `List[torch.Generator]`, *optional*): - A [`torch.Generator`](https://pytorch.org/docs/stable/generated/torch.Generator.html) to make - generation deterministic. - latents (`torch.FloatTensor`, *optional*): - Pre-generated noisy latents sampled from a Gaussian distribution, to be used as inputs for image - generation. Can be used to tweak the same generation with different prompts. If not provided, a latents - tensor is generated by sampling using the supplied random `generator`. - prompt_embeds (`torch.FloatTensor`, *optional*): - Pre-generated text embeddings. Can be used to easily tweak text inputs (prompt weighting). If not - provided, text embeddings are generated from the `prompt` input argument. - negative_prompt_embeds (`torch.FloatTensor`, *optional*): - Pre-generated negative text embeddings. Can be used to easily tweak text inputs (prompt weighting). If - not provided, `negative_prompt_embeds` are generated from the `negative_prompt` input argument. - pooled_prompt_embeds (`torch.FloatTensor`, *optional*): - Pre-generated pooled text embeddings. Can be used to easily tweak text inputs (prompt weighting). If - not provided, pooled text embeddings are generated from `prompt` input argument. - negative_pooled_prompt_embeds (`torch.FloatTensor`, *optional*): - Pre-generated negative pooled text embeddings. Can be used to easily tweak text inputs (prompt - weighting). If not provided, pooled `negative_prompt_embeds` are generated from `negative_prompt` input - argument. - ip_adapter_image: (`PipelineImageInput`, *optional*): Optional image input to work with IP Adapters. - ip_adapter_image_embeds (`List[torch.FloatTensor]`, *optional*): - Pre-generated image embeddings for IP-Adapter. It should be a list of length same as number of IP-adapters. - Each element should be a tensor of shape `(batch_size, num_images, emb_dim)`. It should contain the negative image embedding - if `do_classifier_free_guidance` is set to `True`. - If not provided, embeddings are computed from the `ip_adapter_image` input argument. - output_type (`str`, *optional*, defaults to `"pil"`): - The output format of the generated image. Choose between `PIL.Image` or `np.array`. - return_dict (`bool`, *optional*, defaults to `True`): - Whether or not to return a [`~pipelines.stable_diffusion.StableDiffusionPipelineOutput`] instead of a - plain tuple. - cross_attention_kwargs (`dict`, *optional*): - A kwargs dictionary that if specified is passed along to the [`AttentionProcessor`] as defined in - [`self.processor`](https://github.com/huggingface/diffusers/blob/main/src/diffusers/models/attention_processor.py). - brushnet_conditioning_scale (`float` or `List[float]`, *optional*, defaults to 1.0): - The outputs of the BrushNet are multiplied by `brushnet_conditioning_scale` before they are added - to the residual in the original `unet`. If multiple BrushNets are specified in `init`, you can set - the corresponding scale as a list. - guess_mode (`bool`, *optional*, defaults to `False`): - The BrushNet encoder tries to recognize the content of the input image even if you remove all - prompts. A `guidance_scale` value between 3.0 and 5.0 is recommended. - control_guidance_start (`float` or `List[float]`, *optional*, defaults to 0.0): - The percentage of total steps at which the BrushNet starts applying. - control_guidance_end (`float` or `List[float]`, *optional*, defaults to 1.0): - The percentage of total steps at which the BrushNet stops applying. - original_size (`Tuple[int]`, *optional*, defaults to (1024, 1024)): - If `original_size` is not the same as `target_size` the image will appear to be down- or upsampled. - `original_size` defaults to `(height, width)` if not specified. Part of SDXL's micro-conditioning as - explained in section 2.2 of - [https://huggingface.co/papers/2307.01952](https://huggingface.co/papers/2307.01952). - crops_coords_top_left (`Tuple[int]`, *optional*, defaults to (0, 0)): - `crops_coords_top_left` can be used to generate an image that appears to be "cropped" from the position - `crops_coords_top_left` downwards. Favorable, well-centered images are usually achieved by setting - `crops_coords_top_left` to (0, 0). Part of SDXL's micro-conditioning as explained in section 2.2 of - [https://huggingface.co/papers/2307.01952](https://huggingface.co/papers/2307.01952). - target_size (`Tuple[int]`, *optional*, defaults to (1024, 1024)): - For most cases, `target_size` should be set to the desired height and width of the generated image. If - not specified it will default to `(height, width)`. Part of SDXL's micro-conditioning as explained in - section 2.2 of [https://huggingface.co/papers/2307.01952](https://huggingface.co/papers/2307.01952). - negative_original_size (`Tuple[int]`, *optional*, defaults to (1024, 1024)): - To negatively condition the generation process based on a specific image resolution. Part of SDXL's - micro-conditioning as explained in section 2.2 of - [https://huggingface.co/papers/2307.01952](https://huggingface.co/papers/2307.01952). For more - information, refer to this issue thread: https://github.com/huggingface/diffusers/issues/4208. - negative_crops_coords_top_left (`Tuple[int]`, *optional*, defaults to (0, 0)): - To negatively condition the generation process based on a specific crop coordinates. Part of SDXL's - micro-conditioning as explained in section 2.2 of - [https://huggingface.co/papers/2307.01952](https://huggingface.co/papers/2307.01952). For more - information, refer to this issue thread: https://github.com/huggingface/diffusers/issues/4208. - negative_target_size (`Tuple[int]`, *optional*, defaults to (1024, 1024)): - To negatively condition the generation process based on a target image resolution. It should be as same - as the `target_size` for most cases. Part of SDXL's micro-conditioning as explained in section 2.2 of - [https://huggingface.co/papers/2307.01952](https://huggingface.co/papers/2307.01952). For more - information, refer to this issue thread: https://github.com/huggingface/diffusers/issues/4208. - clip_skip (`int`, *optional*): - Number of layers to be skipped from CLIP while computing the prompt embeddings. A value of 1 means that - the output of the pre-final layer will be used for computing the prompt embeddings. - callback_on_step_end (`Callable`, *optional*): - A function that calls at the end of each denoising steps during the inference. The function is called - with the following arguments: `callback_on_step_end(self: DiffusionPipeline, step: int, timestep: int, - callback_kwargs: Dict)`. `callback_kwargs` will include a list of all tensors as specified by - `callback_on_step_end_tensor_inputs`. - callback_on_step_end_tensor_inputs (`List`, *optional*): - The list of tensor inputs for the `callback_on_step_end` function. The tensors specified in the list - will be passed as `callback_kwargs` argument. You will only be able to include variables listed in the - `._callback_tensor_inputs` attribute of your pipeine class. - - Examples: - - Returns: - [`~pipelines.stable_diffusion.StableDiffusionPipelineOutput`] or `tuple`: - If `return_dict` is `True`, [`~pipelines.stable_diffusion.StableDiffusionPipelineOutput`] is returned, - otherwise a `tuple` is returned containing the output images. - """ - - callback = kwargs.pop("callback", None) - callback_steps = kwargs.pop("callback_steps", None) - - if callback is not None: - deprecate( - "callback", - "1.0.0", - "Passing `callback` as an input argument to `__call__` is deprecated, consider using `callback_on_step_end`", - ) - if callback_steps is not None: - deprecate( - "callback_steps", - "1.0.0", - "Passing `callback_steps` as an input argument to `__call__` is deprecated, consider using `callback_on_step_end`", - ) - - brushnet = self.brushnet._orig_mod if is_compiled_module(self.brushnet) else self.brushnet - - # align format for control guidance - if not isinstance(control_guidance_start, list) and isinstance(control_guidance_end, list): - control_guidance_start = len(control_guidance_end) * [control_guidance_start] - elif not isinstance(control_guidance_end, list) and isinstance(control_guidance_start, list): - control_guidance_end = len(control_guidance_start) * [control_guidance_end] - elif not isinstance(control_guidance_start, list) and not isinstance(control_guidance_end, list): - # mult = len(brushnet.nets) if isinstance(brushnet, MultiBrushNetModel) else 1 - mult = 1 - control_guidance_start, control_guidance_end = ( - mult * [control_guidance_start], - mult * [control_guidance_end], - ) - - # 1. Check inputs. Raise error if not correct - self.check_inputs( - prompt, - prompt_2, - image, - mask, - callback_steps, - negative_prompt, - negative_prompt_2, - prompt_embeds, - negative_prompt_embeds, - pooled_prompt_embeds, - ip_adapter_image, - ip_adapter_image_embeds, - negative_pooled_prompt_embeds, - brushnet_conditioning_scale, - control_guidance_start, - control_guidance_end, - callback_on_step_end_tensor_inputs, - ) - - self._guidance_scale = guidance_scale - self._clip_skip = clip_skip - self._cross_attention_kwargs = cross_attention_kwargs - self._denoising_end = denoising_end - - # 2. Define call parameters - if prompt is not None and isinstance(prompt, str): - batch_size = 1 - elif prompt is not None and isinstance(prompt, list): - batch_size = len(prompt) - else: - batch_size = prompt_embeds.shape[0] - - device = self._execution_device - - # if isinstance(brushnet, MultiBrushNetModel) and isinstance(brushnet_conditioning_scale, float): - # brushnet_conditioning_scale = [brushnet_conditioning_scale] * len(brushnet.nets) - - global_pool_conditions = ( - brushnet.config.global_pool_conditions - if isinstance(brushnet, BrushNetModel) - else brushnet.nets[0].config.global_pool_conditions - ) - guess_mode = guess_mode or global_pool_conditions - - # 3.1 Encode input prompt - text_encoder_lora_scale = ( - self.cross_attention_kwargs.get("scale", None) if self.cross_attention_kwargs is not None else None - ) - ( - prompt_embeds, - negative_prompt_embeds, - pooled_prompt_embeds, - negative_pooled_prompt_embeds, - ) = self.encode_prompt( - prompt, - prompt_2, - device, - num_images_per_prompt, - self.do_classifier_free_guidance, - negative_prompt, - negative_prompt_2, - prompt_embeds=prompt_embeds, - negative_prompt_embeds=negative_prompt_embeds, - pooled_prompt_embeds=pooled_prompt_embeds, - negative_pooled_prompt_embeds=negative_pooled_prompt_embeds, - lora_scale=text_encoder_lora_scale, - clip_skip=self.clip_skip, - ) - - # 3.2 Encode ip_adapter_image - if ip_adapter_image is not None or ip_adapter_image_embeds is not None: - image_embeds = self.prepare_ip_adapter_image_embeds( - ip_adapter_image, - ip_adapter_image_embeds, - device, - batch_size * num_images_per_prompt, - self.do_classifier_free_guidance, - ) - - # 4. Prepare image - if isinstance(brushnet, BrushNetModel): - image = self.prepare_image( - image=image, - width=width, - height=height, - batch_size=batch_size * num_images_per_prompt, - num_images_per_prompt=num_images_per_prompt, - device=device, - dtype=brushnet.dtype, - do_classifier_free_guidance=self.do_classifier_free_guidance, - guess_mode=guess_mode, - ) - original_mask = self.prepare_image( - image=mask, - width=width, - height=height, - batch_size=batch_size * num_images_per_prompt, - num_images_per_prompt=num_images_per_prompt, - device=device, - dtype=brushnet.dtype, - do_classifier_free_guidance=self.do_classifier_free_guidance, - guess_mode=guess_mode, - ) - original_mask=(original_mask.sum(1)[:,None,:,:] < 0).to(image.dtype) - height, width = image.shape[-2:] - # elif isinstance(brushnet, MultiBrushNetModel): - # images = [] - - # for image_ in image: - # image_ = self.prepare_image( - # image=image_, - # width=width, - # height=height, - # batch_size=batch_size * num_images_per_prompt, - # num_images_per_prompt=num_images_per_prompt, - # device=device, - # dtype=brushnet.dtype, - # do_classifier_free_guidance=self.do_classifier_free_guidance, - # guess_mode=guess_mode, - # ) - - # images.append(image_) - - # image = images - # height, width = image[0].shape[-2:] - else: - assert False - - # 5. Prepare timesteps - self.scheduler.set_timesteps(num_inference_steps, device=device) - timesteps = self.scheduler.timesteps - self._num_timesteps = len(timesteps) - - # 6. Prepare latent variables - num_channels_latents = self.unet.config.in_channels - latents, noise = self.prepare_latents( - batch_size * num_images_per_prompt, - num_channels_latents, - height, - width, - prompt_embeds.dtype, - device, - generator, - latents, - ) - - # 6.1 prepare condition latents - conditioning_latents=self.vae.encode(image).latent_dist.sample() * self.vae.config.scaling_factor - mask = torch.nn.functional.interpolate( - original_mask, - size=( - conditioning_latents.shape[-2], - conditioning_latents.shape[-1] - ) - ) - conditioning_latents = torch.concat([conditioning_latents,mask],1) - - - # 6.5 Optionally get Guidance Scale Embedding - timestep_cond = None - if self.unet.config.time_cond_proj_dim is not None: - guidance_scale_tensor = torch.tensor(self.guidance_scale - 1).repeat(batch_size * num_images_per_prompt) - timestep_cond = self.get_guidance_scale_embedding( - guidance_scale_tensor, embedding_dim=self.unet.config.time_cond_proj_dim - ).to(device=device, dtype=latents.dtype) - - # 7. Prepare extra step kwargs. TODO: Logic should ideally just be moved out of the pipeline - extra_step_kwargs = self.prepare_extra_step_kwargs(generator, eta) - - # 7.1 Create tensor stating which brushnets to keep - brushnet_keep = [] - for i in range(len(timesteps)): - keeps = [ - 1.0 - float(i / len(timesteps) < s or (i + 1) / len(timesteps) > e) - for s, e in zip(control_guidance_start, control_guidance_end) - ] - brushnet_keep.append(keeps[0] if isinstance(brushnet, BrushNetModel) else keeps) - - # 7.2 Prepare added time ids & embeddings - if isinstance(image, list): - original_size = original_size or image[0].shape[-2:] - else: - original_size = original_size or image.shape[-2:] - target_size = target_size or (height, width) - - add_text_embeds = pooled_prompt_embeds - if self.text_encoder_2 is None: - text_encoder_projection_dim = int(pooled_prompt_embeds.shape[-1]) - else: - text_encoder_projection_dim = self.text_encoder_2.config.projection_dim - - add_time_ids = self._get_add_time_ids( - original_size, - crops_coords_top_left, - target_size, - dtype=prompt_embeds.dtype, - text_encoder_projection_dim=text_encoder_projection_dim, - ) - - if negative_original_size is not None and negative_target_size is not None: - negative_add_time_ids = self._get_add_time_ids( - negative_original_size, - negative_crops_coords_top_left, - negative_target_size, - dtype=prompt_embeds.dtype, - text_encoder_projection_dim=text_encoder_projection_dim, - ) - else: - negative_add_time_ids = add_time_ids - - if self.do_classifier_free_guidance: - prompt_embeds = torch.cat([negative_prompt_embeds, prompt_embeds], dim=0) - add_text_embeds = torch.cat([negative_pooled_prompt_embeds, add_text_embeds], dim=0) - add_time_ids = torch.cat([negative_add_time_ids, add_time_ids], dim=0) - - prompt_embeds = prompt_embeds.to(device) - add_text_embeds = add_text_embeds.to(device) - add_time_ids = add_time_ids.to(device).repeat(batch_size * num_images_per_prompt, 1) - - # 8. Denoising loop - num_warmup_steps = len(timesteps) - num_inference_steps * self.scheduler.order - - # 8.1 Apply denoising_end - if ( - self.denoising_end is not None - and isinstance(self.denoising_end, float) - and self.denoising_end > 0 - and self.denoising_end < 1 - ): - discrete_timestep_cutoff = int( - round( - self.scheduler.config.num_train_timesteps - - (self.denoising_end * self.scheduler.config.num_train_timesteps) - ) - ) - num_inference_steps = len(list(filter(lambda ts: ts >= discrete_timestep_cutoff, timesteps))) - timesteps = timesteps[:num_inference_steps] - - is_unet_compiled = is_compiled_module(self.unet) - is_brushnet_compiled = is_compiled_module(self.brushnet) - is_torch_higher_equal_2_1 = is_torch_version(">=", "2.1") - with self.progress_bar(total=num_inference_steps) as progress_bar: - for i, t in enumerate(timesteps): - # Relevant thread: - # https://dev-discuss.pytorch.org/t/cudagraphs-in-pytorch-2-0/1428 - if (is_unet_compiled and is_brushnet_compiled) and is_torch_higher_equal_2_1: - torch._inductor.cudagraph_mark_step_begin() - # expand the latents if we are doing classifier free guidance - latent_model_input = torch.cat([latents] * 2) if self.do_classifier_free_guidance else latents - latent_model_input = self.scheduler.scale_model_input(latent_model_input, t) - - added_cond_kwargs = {"text_embeds": add_text_embeds, "time_ids": add_time_ids} - - # brushnet(s) inference - if guess_mode and self.do_classifier_free_guidance: - # Infer BrushNet only for the conditional batch. - control_model_input = latents - control_model_input = self.scheduler.scale_model_input(control_model_input, t) - brushnet_prompt_embeds = prompt_embeds.chunk(2)[1] - brushnet_added_cond_kwargs = { - "text_embeds": add_text_embeds.chunk(2)[1], - "time_ids": add_time_ids.chunk(2)[1], - } - else: - control_model_input = latent_model_input - brushnet_prompt_embeds = prompt_embeds - brushnet_added_cond_kwargs = added_cond_kwargs - - if isinstance(brushnet_keep[i], list): - cond_scale = [c * s for c, s in zip(brushnet_conditioning_scale, brushnet_keep[i])] - else: - brushnet_cond_scale = brushnet_conditioning_scale - if isinstance(brushnet_cond_scale, list): - brushnet_cond_scale = brushnet_cond_scale[0] - cond_scale = brushnet_cond_scale * brushnet_keep[i] - - down_block_res_samples, mid_block_res_sample, up_block_res_samples = self.brushnet( - sample=control_model_input, - encoder_hidden_states=brushnet_prompt_embeds, - brushnet_cond=conditioning_latents, - timestep=t, - conditioning_scale=cond_scale, - guess_mode=guess_mode, - added_cond_kwargs=brushnet_added_cond_kwargs, - return_dict=False, - ) - - if guess_mode and self.do_classifier_free_guidance: - # Infered BrushNet only for the conditional batch. - # To apply the output of BrushNet to both the unconditional and conditional batches, - # add 0 to the unconditional batch to keep it unchanged. - down_block_res_samples = [torch.cat([torch.zeros_like(d), d]) for d in down_block_res_samples] - mid_block_res_sample = torch.cat([torch.zeros_like(mid_block_res_sample), mid_block_res_sample]) - up_block_res_samples = [torch.cat([torch.zeros_like(d), d]) for d in up_block_res_samples] - - if ip_adapter_image is not None or ip_adapter_image_embeds is not None: - added_cond_kwargs["image_embeds"] = image_embeds - - # predict the noise residual - noise_pred = self.unet( - latent_model_input, - t, - encoder_hidden_states=prompt_embeds, - timestep_cond=timestep_cond, - cross_attention_kwargs=self.cross_attention_kwargs, - down_block_add_samples=down_block_res_samples, - mid_block_add_sample=mid_block_res_sample, - up_block_add_samples=up_block_res_samples, - added_cond_kwargs=added_cond_kwargs, - return_dict=False, - )[0] - - # perform guidance - if self.do_classifier_free_guidance: - noise_pred_uncond, noise_pred_text = noise_pred.chunk(2) - noise_pred = noise_pred_uncond + guidance_scale * (noise_pred_text - noise_pred_uncond) - - # compute the previous noisy sample x_t -> x_t-1 - latents = self.scheduler.step(noise_pred, t, latents, **extra_step_kwargs, return_dict=False)[0] - - if callback_on_step_end is not None: - callback_kwargs = {} - for k in callback_on_step_end_tensor_inputs: - callback_kwargs[k] = locals()[k] - callback_outputs = callback_on_step_end(self, i, t, callback_kwargs) - - latents = callback_outputs.pop("latents", latents) - prompt_embeds = callback_outputs.pop("prompt_embeds", prompt_embeds) - negative_prompt_embeds = callback_outputs.pop("negative_prompt_embeds", negative_prompt_embeds) - - # call the callback, if provided - if i == len(timesteps) - 1 or ((i + 1) > num_warmup_steps and (i + 1) % self.scheduler.order == 0): - progress_bar.update() - if callback is not None and i % callback_steps == 0: - step_idx = i // getattr(self.scheduler, "order", 1) - callback(step_idx, t, latents) - - if not output_type == "latent": - # make sure the VAE is in float32 mode, as it overflows in float16 - needs_upcasting = self.vae.dtype == torch.float16 and self.vae.config.force_upcast - - if needs_upcasting: - self.upcast_vae() - latents = latents.to(next(iter(self.vae.post_quant_conv.parameters())).dtype) - - # unscale/denormalize the latents - # denormalize with the mean and std if available and not None - has_latents_mean = hasattr(self.vae.config, "latents_mean") and self.vae.config.latents_mean is not None - has_latents_std = hasattr(self.vae.config, "latents_std") and self.vae.config.latents_std is not None - if has_latents_mean and has_latents_std: - latents_mean = ( - torch.tensor(self.vae.config.latents_mean).view(1, 4, 1, 1).to(latents.device, latents.dtype) - ) - latents_std = ( - torch.tensor(self.vae.config.latents_std).view(1, 4, 1, 1).to(latents.device, latents.dtype) - ) - latents = latents * latents_std / self.vae.config.scaling_factor + latents_mean - else: - latents = latents / self.vae.config.scaling_factor - - image = self.vae.decode(latents, return_dict=False)[0] - - # cast back to fp16 if needed - if needs_upcasting: - self.vae.to(dtype=torch.float16) - else: - image = latents - - if not output_type == "latent": - # apply watermark if available - if self.watermark is not None: - image = self.watermark.apply_watermark(image) - - image = self.image_processor.postprocess(image, output_type=output_type) - - # Offload all models - self.maybe_free_model_hooks() - - if not return_dict: - return (image,) - - return StableDiffusionXLPipelineOutput(images=image) diff --git a/brushnet/pipeline_brushnet_powerpaint_ca.py b/brushnet/pipeline_powerpaint_brushnet_ca.py similarity index 96% rename from brushnet/pipeline_brushnet_powerpaint_ca.py rename to brushnet/pipeline_powerpaint_brushnet_ca.py index 5a2286f..88f7b7f 100644 --- a/brushnet/pipeline_brushnet_powerpaint_ca.py +++ b/brushnet/pipeline_powerpaint_brushnet_ca.py @@ -1,18 +1,25 @@ import inspect -from typing import Any, Callable, Dict, List, Optional, Tuple, Union +import sys +from typing import Any, Callable, Dict, List, Optional, Union import numpy as np import PIL.Image import torch import torch.nn.functional as F + + +sys.path.append("..") from transformers import CLIPImageProcessor, CLIPTextModel, CLIPTokenizer, CLIPVisionModelWithProjection -from ...image_processor import PipelineImageInput, VaeImageProcessor -from ...loaders import FromSingleFileMixin, IPAdapterMixin, LoraLoaderMixin, TextualInversionLoaderMixin -from ...models import AutoencoderKL, BrushNetCAModel, ImageProjection, UNet2DConditionModel -from ...models.lora import adjust_lora_scale_text_encoder -from ...schedulers import KarrasDiffusionSchedulers -from ...utils import ( +from diffusers.image_processor import PipelineImageInput, VaeImageProcessor +from diffusers.loaders import FromSingleFileMixin, IPAdapterMixin, LoraLoaderMixin, TextualInversionLoaderMixin +from diffusers.models import AutoencoderKL +from diffusers.models.lora import adjust_lora_scale_text_encoder +from diffusers.pipelines.pipeline_utils import DiffusionPipeline +from diffusers.pipelines.stable_diffusion.pipeline_output import StableDiffusionPipelineOutput +from diffusers.pipelines.stable_diffusion.safety_checker import StableDiffusionSafetyChecker +from diffusers.schedulers import KarrasDiffusionSchedulers +from diffusers.utils import ( USE_PEFT_BACKEND, deprecate, logging, @@ -20,10 +27,12 @@ from ...utils import ( scale_lora_layers, unscale_lora_layers, ) -from ...utils.torch_utils import is_compiled_module, is_torch_version, randn_tensor -from ..pipeline_utils import DiffusionPipeline, StableDiffusionMixin -from ..stable_diffusion.pipeline_output import StableDiffusionPipelineOutput -from ..stable_diffusion.safety_checker import StableDiffusionSafetyChecker +from diffusers.utils.torch_utils import is_compiled_module, is_torch_version, randn_tensor + +from .unet_2d_condition import UNet2DConditionModel +from .brushnet_ca import BrushNetModel + +from diffusers.pipelines.pipeline_utils import StableDiffusionMixin logger = logging.get_logger(__name__) # pylint: disable=invalid-name @@ -68,10 +77,10 @@ EXAMPLE_DOC_STRING = """ generator = torch.Generator("cuda").manual_seed(1234) image = pipe( - caption, - init_image, - mask_image, - num_inference_steps=50, + caption, + init_image, + mask_image, + num_inference_steps=50, generator=generator, paintingnet_conditioning_scale=1.0 ).images[0] @@ -125,7 +134,7 @@ def retrieve_timesteps( return timesteps, num_inference_steps -class StableDiffusionBrushNetPowerPaintCAPipeline( +class StableDiffusionPowerPaintBrushNetPipeline( DiffusionPipeline, StableDiffusionMixin, TextualInversionLoaderMixin, @@ -180,7 +189,7 @@ class StableDiffusionBrushNetPowerPaintCAPipeline( text_encoder_brushnet: CLIPTextModel, tokenizer: CLIPTokenizer, unet: UNet2DConditionModel, - brushnet: BrushNetCAModel, + brushnet: BrushNetModel, scheduler: KarrasDiffusionSchedulers, safety_checker: StableDiffusionSafetyChecker, feature_extractor: CLIPImageProcessor, @@ -208,7 +217,7 @@ class StableDiffusionBrushNetPowerPaintCAPipeline( self.register_modules( vae=vae, text_encoder=text_encoder, - text_encoder_brushnet = text_encoder_brushnet, + text_encoder_brushnet=text_encoder_brushnet, tokenizer=tokenizer, unet=unet, brushnet=brushnet, @@ -232,14 +241,11 @@ class StableDiffusionBrushNetPowerPaintCAPipeline( do_classifier_free_guidance, negative_promptA=None, negative_promptB=None, - t_nag = None, + t_nag=None, prompt_embeds: Optional[torch.FloatTensor] = None, negative_prompt_embeds: Optional[torch.FloatTensor] = None, lora_scale: Optional[float] = None, ): - deprecation_message = "`_encode_prompt()` is deprecated and it will be removed in a future version. Use `encode_prompt()` instead. Also, be aware that the output format changed from a concatenated tensor to a tuple." - deprecate("_encode_prompt()", "1.0.0", deprecation_message, standard_warn=False) - r""" Encodes the prompt into text encoder hidden states. @@ -270,7 +276,7 @@ class StableDiffusionBrushNetPowerPaintCAPipeline( # function of text encoder can correctly access it if lora_scale is not None and isinstance(self, LoraLoaderMixin): self._lora_scale = lora_scale - + prompt = promptA negative_prompt = negative_promptA @@ -315,10 +321,14 @@ class StableDiffusionBrushNetPowerPaintCAPipeline( f" {self.tokenizer.model_max_length} tokens: {removed_text}" ) - if hasattr(self.text_encoder_brushnet.config, "use_attention_mask") and self.text_encoder_brushnet.config.use_attention_mask: - attention_mask = text_inputsA.attention_mask.to(device) - else: - attention_mask = None + ### NQ's removal + #if ( + # hasattr(self.text_encoder_brushnet.config, "use_attention_mask") + # and self.text_encoder_brushnet.config.use_attention_mask + #): + # attention_mask = text_inputsA.attention_mask.to(device) + #else: + attention_mask = None # print("text_input_idsA: ",text_input_idsA) # print("text_input_idsB: ",text_input_idsB) @@ -335,7 +345,7 @@ class StableDiffusionBrushNetPowerPaintCAPipeline( attention_mask=attention_mask, ) prompt_embedsB = prompt_embedsB[0] - prompt_embeds = prompt_embedsA*(t)+(1-t)*prompt_embedsB + prompt_embeds = prompt_embedsA * (t) + (1 - t) * prompt_embedsB # print("prompt_embeds: ",prompt_embeds) if self.text_encoder_brushnet is not None: @@ -398,10 +408,14 @@ class StableDiffusionBrushNetPowerPaintCAPipeline( return_tensors="pt", ) - if hasattr(self.text_encoder_brushnet.config, "use_attention_mask") and self.text_encoder_brushnet.config.use_attention_mask: - attention_mask = uncond_inputA.attention_mask.to(device) - else: - attention_mask = None + ### NQ's remove + #if ( + # hasattr(self.text_encoder_brushnet.config, "use_attention_mask") + # and self.text_encoder_brushnet.config.use_attention_mask + #): + # attention_mask = uncond_inputA.attention_mask.to(device) + #else: + attention_mask = None negative_prompt_embedsA = self.text_encoder_brushnet( uncond_inputA.input_ids.to(device), @@ -411,7 +425,7 @@ class StableDiffusionBrushNetPowerPaintCAPipeline( uncond_inputB.input_ids.to(device), attention_mask=attention_mask, ) - negative_prompt_embeds = negative_prompt_embedsA[0]*(t_nag)+(1-t_nag)*negative_prompt_embedsB[0] + negative_prompt_embeds = negative_prompt_embedsA[0] * (t_nag) + (1 - t_nag) * negative_prompt_embedsB[0] # negative_prompt_embeds = negative_prompt_embeds[0] @@ -804,9 +818,9 @@ class StableDiffusionBrushNetPowerPaintCAPipeline( self.brushnet, torch._dynamo.eval_frame.OptimizedModule ) if ( - isinstance(self.brushnet, BrushNetCAModel) + isinstance(self.brushnet, BrushNetModel) or is_compiled - and isinstance(self.brushnet._orig_mod, BrushNetCAModel) + and isinstance(self.brushnet._orig_mod, BrushNetModel) ): self.check_image(image, mask, prompt, prompt_embeds) else: @@ -814,9 +828,9 @@ class StableDiffusionBrushNetPowerPaintCAPipeline( # Check `brushnet_conditioning_scale` if ( - isinstance(self.brushnet, BrushNetCAModel) + isinstance(self.brushnet, BrushNetModel) or is_compiled - and isinstance(self.brushnet._orig_mod, BrushNetCAModel) + and isinstance(self.brushnet._orig_mod, BrushNetModel) ): if not isinstance(brushnet_conditioning_scale, float): raise TypeError("For single brushnet: `brushnet_conditioning_scale` must be type `float`.") @@ -1233,7 +1247,7 @@ class StableDiffusionBrushNetPowerPaintCAPipeline( global_pool_conditions = ( brushnet.config.global_pool_conditions - if isinstance(brushnet, BrushNetCAModel) + if isinstance(brushnet, BrushNetModel) else brushnet.nets[0].config.global_pool_conditions ) guess_mode = guess_mode or global_pool_conditions @@ -1242,6 +1256,7 @@ class StableDiffusionBrushNetPowerPaintCAPipeline( text_encoder_lora_scale = ( self.cross_attention_kwargs.get("scale", None) if self.cross_attention_kwargs is not None else None ) + prompt_embeds = self._encode_prompt( promptA, promptB, @@ -1269,8 +1284,6 @@ class StableDiffusionBrushNetPowerPaintCAPipeline( lora_scale=text_encoder_lora_scale, ) - - if ip_adapter_image is not None or ip_adapter_image_embeds is not None: image_embeds = self.prepare_ip_adapter_image_embeds( ip_adapter_image, @@ -1281,7 +1294,7 @@ class StableDiffusionBrushNetPowerPaintCAPipeline( ) # 4. Prepare image - if isinstance(brushnet, BrushNetCAModel): + if isinstance(brushnet, BrushNetModel): image = self.prepare_image( image=image, width=width, @@ -1304,7 +1317,7 @@ class StableDiffusionBrushNetPowerPaintCAPipeline( do_classifier_free_guidance=self.do_classifier_free_guidance, guess_mode=guess_mode, ) - original_mask=(original_mask.sum(1)[:,None,:,:] < 0).to(image.dtype) + original_mask = (original_mask.sum(1)[:, None, :, :] < 0).to(image.dtype) height, width = image.shape[-2:] else: assert False @@ -1327,25 +1340,22 @@ class StableDiffusionBrushNetPowerPaintCAPipeline( ) # 6.1 prepare condition latents - from torchvision import transforms # mask_i = transforms.ToPILImage()(image[0:1,:,:,:].squeeze(0)) # mask_i.save('_mask.png') # print(brushnet.dtype) - conditioning_latents=self.vae.encode(image.to(device=device, dtype=brushnet.dtype)).latent_dist.sample() * self.vae.config.scaling_factor + conditioning_latents = ( + self.vae.encode(image.to(device=device, dtype=brushnet.dtype)).latent_dist.sample() + * self.vae.config.scaling_factor + ) mask = torch.nn.functional.interpolate( - original_mask, - size=( - conditioning_latents.shape[-2], - conditioning_latents.shape[-1] - ) - ) - conditioning_latents = torch.concat([conditioning_latents,mask],1) + original_mask, size=(conditioning_latents.shape[-2], conditioning_latents.shape[-1]) + ) + conditioning_latents = torch.concat([conditioning_latents, mask], 1) # image = self.vae.decode(conditioning_latents[:1,:4,:,:] / self.vae.config.scaling_factor, return_dict=False, generator=generator)[0] # from torchvision import transforms # mask_i = transforms.ToPILImage()(image[0:1,:,:,:].squeeze(0)/2+0.5) # mask_i.save(str(timesteps[0]) +'_C.png') - # 6.5 Optionally get Guidance Scale Embedding timestep_cond = None if self.unet.config.time_cond_proj_dim is not None: @@ -1371,7 +1381,7 @@ class StableDiffusionBrushNetPowerPaintCAPipeline( 1.0 - float(i / len(timesteps) < s or (i + 1) / len(timesteps) > e) for s, e in zip(control_guidance_start, control_guidance_end) ] - brushnet_keep.append(keeps[0] if isinstance(brushnet, BrushNetCAModel) else keeps) + brushnet_keep.append(keeps[0] if isinstance(brushnet, BrushNetModel) else keeps) # 8. Denoising loop num_warmup_steps = len(timesteps) - num_inference_steps * self.scheduler.order diff --git a/brushnet/powerpaint.json b/brushnet/powerpaint.json new file mode 100644 index 0000000..4d7c73e --- /dev/null +++ b/brushnet/powerpaint.json @@ -0,0 +1,57 @@ +{ + "_class_name": "BrushNetModel", + "_diffusers_version": "0.27.2", + "act_fn": "silu", + "addition_embed_type": null, + "addition_embed_type_num_heads": 64, + "addition_time_embed_dim": null, + "attention_head_dim": 8, + "block_out_channels": [ + 320, + 640, + 1280, + 1280 + ], + "brushnet_conditioning_channel_order": "rgb", + "class_embed_type": null, + "conditioning_channels": 5, + "conditioning_embedding_out_channels": [ + 16, + 32, + 96, + 256 + ], + "cross_attention_dim": 768, + "down_block_types": [ + "CrossAttnDownBlock2D", + "CrossAttnDownBlock2D", + "CrossAttnDownBlock2D", + "DownBlock2D" + ], + "downsample_padding": 1, + "encoder_hid_dim": null, + "encoder_hid_dim_type": null, + "flip_sin_to_cos": true, + "freq_shift": 0, + "global_pool_conditions": false, + "in_channels": 4, + "layers_per_block": 2, + "mid_block_scale_factor": 1, + "mid_block_type": "UNetMidBlock2DCrossAttn", + "norm_eps": 1e-05, + "norm_num_groups": 32, + "num_attention_heads": null, + "num_class_embeds": null, + "only_cross_attention": false, + "projection_class_embeddings_input_dim": null, + "resnet_time_scale_shift": "default", + "transformer_layers_per_block": 1, + "up_block_types": [ + "UpBlock2D", + "CrossAttnUpBlock2D", + "CrossAttnUpBlock2D", + "CrossAttnUpBlock2D" + ], + "upcast_attention": false, + "use_linear_projection": false +} diff --git a/brushnet/powerpaint_utils.py b/brushnet/powerpaint_utils.py new file mode 100644 index 0000000..bcbb1f7 --- /dev/null +++ b/brushnet/powerpaint_utils.py @@ -0,0 +1,496 @@ +import copy +import random + +import torch +import torch.nn as nn +from transformers import CLIPTokenizer +from typing import Any, List, Optional, Union + +class TokenizerWrapper: + """Tokenizer wrapper for CLIPTokenizer. Only support CLIPTokenizer + currently. This wrapper is modified from https://github.com/huggingface/dif + fusers/blob/e51f19aee82c8dd874b715a09dbc521d88835d68/src/diffusers/loaders. + py#L358 # noqa. + + Args: + from_pretrained (Union[str, os.PathLike], optional): The *model id* + of a pretrained model or a path to a *directory* containing + model weights and config. Defaults to None. + from_config (Union[str, os.PathLike], optional): The *model id* + of a pretrained model or a path to a *directory* containing + model weights and config. Defaults to None. + + *args, **kwargs: If `from_pretrained` is passed, *args and **kwargs + will be passed to `from_pretrained` function. Otherwise, *args + and **kwargs will be used to initialize the model by + `self._module_cls(*args, **kwargs)`. + """ + + def __init__(self, tokenizer: CLIPTokenizer): + self.wrapped = tokenizer + self.token_map = {} + + def __getattr__(self, name: str) -> Any: + if name in self.__dict__: + return getattr(self, name) + #if name == "wrapped": + # return getattr(self, 'wrapped')#super().__getattr__("wrapped") + + try: + return getattr(self.wrapped, name) + except AttributeError: + raise AttributeError( + "'name' cannot be found in both " + f"'{self.__class__.__name__}' and " + f"'{self.__class__.__name__}.tokenizer'." + ) + + def try_adding_tokens(self, tokens: Union[str, List[str]], *args, **kwargs): + """Attempt to add tokens to the tokenizer. + + Args: + tokens (Union[str, List[str]]): The tokens to be added. + """ + num_added_tokens = self.wrapped.add_tokens(tokens, *args, **kwargs) + assert num_added_tokens != 0, ( + f"The tokenizer already contains the token {tokens}. Please pass " + "a different `placeholder_token` that is not already in the " + "tokenizer." + ) + + def get_token_info(self, token: str) -> dict: + """Get the information of a token, including its start and end index in + the current tokenizer. + + Args: + token (str): The token to be queried. + + Returns: + dict: The information of the token, including its start and end + index in current tokenizer. + """ + token_ids = self.__call__(token).input_ids + start, end = token_ids[1], token_ids[-2] + 1 + return {"name": token, "start": start, "end": end} + + def add_placeholder_token(self, placeholder_token: str, *args, num_vec_per_token: int = 1, **kwargs): + """Add placeholder tokens to the tokenizer. + + Args: + placeholder_token (str): The placeholder token to be added. + num_vec_per_token (int, optional): The number of vectors of + the added placeholder token. + *args, **kwargs: The arguments for `self.wrapped.add_tokens`. + """ + output = [] + if num_vec_per_token == 1: + self.try_adding_tokens(placeholder_token, *args, **kwargs) + output.append(placeholder_token) + else: + output = [] + for i in range(num_vec_per_token): + ith_token = placeholder_token + f"_{i}" + self.try_adding_tokens(ith_token, *args, **kwargs) + output.append(ith_token) + + for token in self.token_map: + if token in placeholder_token: + raise ValueError( + f"The tokenizer already has placeholder token {token} " + f"that can get confused with {placeholder_token} " + "keep placeholder tokens independent" + ) + self.token_map[placeholder_token] = output + + def replace_placeholder_tokens_in_text( + self, text: Union[str, List[str]], vector_shuffle: bool = False, prop_tokens_to_load: float = 1.0 + ) -> Union[str, List[str]]: + """Replace the keywords in text with placeholder tokens. This function + will be called in `self.__call__` and `self.encode`. + + Args: + text (Union[str, List[str]]): The text to be processed. + vector_shuffle (bool, optional): Whether to shuffle the vectors. + Defaults to False. + prop_tokens_to_load (float, optional): The proportion of tokens to + be loaded. If 1.0, all tokens will be loaded. Defaults to 1.0. + + Returns: + Union[str, List[str]]: The processed text. + """ + if isinstance(text, list): + output = [] + for i in range(len(text)): + output.append(self.replace_placeholder_tokens_in_text(text[i], vector_shuffle=vector_shuffle)) + return output + + for placeholder_token in self.token_map: + if placeholder_token in text: + tokens = self.token_map[placeholder_token] + tokens = tokens[: 1 + int(len(tokens) * prop_tokens_to_load)] + if vector_shuffle: + tokens = copy.copy(tokens) + random.shuffle(tokens) + text = text.replace(placeholder_token, " ".join(tokens)) + return text + + def replace_text_with_placeholder_tokens(self, text: Union[str, List[str]]) -> Union[str, List[str]]: + """Replace the placeholder tokens in text with the original keywords. + This function will be called in `self.decode`. + + Args: + text (Union[str, List[str]]): The text to be processed. + + Returns: + Union[str, List[str]]: The processed text. + """ + if isinstance(text, list): + output = [] + for i in range(len(text)): + output.append(self.replace_text_with_placeholder_tokens(text[i])) + return output + + for placeholder_token, tokens in self.token_map.items(): + merged_tokens = " ".join(tokens) + if merged_tokens in text: + text = text.replace(merged_tokens, placeholder_token) + return text + + def __call__( + self, + text: Union[str, List[str]], + *args, + vector_shuffle: bool = False, + prop_tokens_to_load: float = 1.0, + **kwargs, + ): + """The call function of the wrapper. + + Args: + text (Union[str, List[str]]): The text to be tokenized. + vector_shuffle (bool, optional): Whether to shuffle the vectors. + Defaults to False. + prop_tokens_to_load (float, optional): The proportion of tokens to + be loaded. If 1.0, all tokens will be loaded. Defaults to 1.0 + *args, **kwargs: The arguments for `self.wrapped.__call__`. + """ + replaced_text = self.replace_placeholder_tokens_in_text( + text, vector_shuffle=vector_shuffle, prop_tokens_to_load=prop_tokens_to_load + ) + + return self.wrapped.__call__(replaced_text, *args, **kwargs) + + def encode(self, text: Union[str, List[str]], *args, **kwargs): + """Encode the passed text to token index. + + Args: + text (Union[str, List[str]]): The text to be encode. + *args, **kwargs: The arguments for `self.wrapped.__call__`. + """ + replaced_text = self.replace_placeholder_tokens_in_text(text) + return self.wrapped(replaced_text, *args, **kwargs) + + def decode(self, token_ids, return_raw: bool = False, *args, **kwargs) -> Union[str, List[str]]: + """Decode the token index to text. + + Args: + token_ids: The token index to be decoded. + return_raw: Whether keep the placeholder token in the text. + Defaults to False. + *args, **kwargs: The arguments for `self.wrapped.decode`. + + Returns: + Union[str, List[str]]: The decoded text. + """ + text = self.wrapped.decode(token_ids, *args, **kwargs) + if return_raw: + return text + replaced_text = self.replace_text_with_placeholder_tokens(text) + return replaced_text + + def __repr__(self): + """The representation of the wrapper.""" + s = super().__repr__() + prefix = f"Wrapped Module Class: {self._module_cls}\n" + prefix += f"Wrapped Module Name: {self._module_name}\n" + if self._from_pretrained: + prefix += f"From Pretrained: {self._from_pretrained}\n" + s = prefix + s + return s + + +class EmbeddingLayerWithFixes(nn.Module): + """The revised embedding layer to support external embeddings. This design + of this class is inspired by https://github.com/AUTOMATIC1111/stable- + diffusion-webui/blob/22bcc7be428c94e9408f589966c2040187245d81/modules/sd_hi + jack.py#L224 # noqa. + + Args: + wrapped (nn.Emebdding): The embedding layer to be wrapped. + external_embeddings (Union[dict, List[dict]], optional): The external + embeddings added to this layer. Defaults to None. + """ + + def __init__(self, wrapped: nn.Embedding, external_embeddings: Optional[Union[dict, List[dict]]] = None): + super().__init__() + self.wrapped = wrapped + self.num_embeddings = wrapped.weight.shape[0] + + self.external_embeddings = [] + if external_embeddings: + self.add_embeddings(external_embeddings) + + self.trainable_embeddings = nn.ParameterDict() + + @property + def weight(self): + """Get the weight of wrapped embedding layer.""" + return self.wrapped.weight + + def check_duplicate_names(self, embeddings: List[dict]): + """Check whether duplicate names exist in list of 'external + embeddings'. + + Args: + embeddings (List[dict]): A list of embedding to be check. + """ + names = [emb["name"] for emb in embeddings] + assert len(names) == len(set(names)), ( + "Found duplicated names in 'external_embeddings'. Name list: " f"'{names}'" + ) + + def check_ids_overlap(self, embeddings): + """Check whether overlap exist in token ids of 'external_embeddings'. + + Args: + embeddings (List[dict]): A list of embedding to be check. + """ + ids_range = [[emb["start"], emb["end"], emb["name"]] for emb in embeddings] + ids_range.sort() # sort by 'start' + # check if 'end' has overlapping + for idx in range(len(ids_range) - 1): + name1, name2 = ids_range[idx][-1], ids_range[idx + 1][-1] + assert ids_range[idx][1] <= ids_range[idx + 1][0], ( + f"Found ids overlapping between embeddings '{name1}' " f"and '{name2}'." + ) + + def add_embeddings(self, embeddings: Optional[Union[dict, List[dict]]]): + """Add external embeddings to this layer. + + Use case: + + >>> 1. Add token to tokenizer and get the token id. + >>> tokenizer = TokenizerWrapper('openai/clip-vit-base-patch32') + >>> # 'how much' in kiswahili + >>> tokenizer.add_placeholder_tokens('ngapi', num_vec_per_token=4) + >>> + >>> 2. Add external embeddings to the model. + >>> new_embedding = { + >>> 'name': 'ngapi', # 'how much' in kiswahili + >>> 'embedding': torch.ones(1, 15) * 4, + >>> 'start': tokenizer.get_token_info('kwaheri')['start'], + >>> 'end': tokenizer.get_token_info('kwaheri')['end'], + >>> 'trainable': False # if True, will registry as a parameter + >>> } + >>> embedding_layer = nn.Embedding(10, 15) + >>> embedding_layer_wrapper = EmbeddingLayerWithFixes(embedding_layer) + >>> embedding_layer_wrapper.add_embeddings(new_embedding) + >>> + >>> 3. Forward tokenizer and embedding layer! + >>> input_text = ['hello, ngapi!', 'hello my friend, ngapi?'] + >>> input_ids = tokenizer( + >>> input_text, padding='max_length', truncation=True, + >>> return_tensors='pt')['input_ids'] + >>> out_feat = embedding_layer_wrapper(input_ids) + >>> + >>> 4. Let's validate the result! + >>> assert (out_feat[0, 3: 7] == 2.3).all() + >>> assert (out_feat[2, 5: 9] == 2.3).all() + + Args: + embeddings (Union[dict, list[dict]]): The external embeddings to + be added. Each dict must contain the following 4 fields: 'name' + (the name of this embedding), 'embedding' (the embedding + tensor), 'start' (the start token id of this embedding), 'end' + (the end token id of this embedding). For example: + `{name: NAME, start: START, end: END, embedding: torch.Tensor}` + """ + if isinstance(embeddings, dict): + embeddings = [embeddings] + + self.external_embeddings += embeddings + self.check_duplicate_names(self.external_embeddings) + self.check_ids_overlap(self.external_embeddings) + + # set for trainable + added_trainable_emb_info = [] + for embedding in embeddings: + trainable = embedding.get("trainable", False) + if trainable: + name = embedding["name"] + embedding["embedding"] = torch.nn.Parameter(embedding["embedding"]) + self.trainable_embeddings[name] = embedding["embedding"] + added_trainable_emb_info.append(name) + + added_emb_info = [emb["name"] for emb in embeddings] + added_emb_info = ", ".join(added_emb_info) + print(f"Successfully add external embeddings: {added_emb_info}.", "current") + + if added_trainable_emb_info: + added_trainable_emb_info = ", ".join(added_trainable_emb_info) + print("Successfully add trainable external embeddings: " f"{added_trainable_emb_info}", "current") + + def replace_input_ids(self, input_ids: torch.Tensor) -> torch.Tensor: + """Replace external input ids to 0. + + Args: + input_ids (torch.Tensor): The input ids to be replaced. + + Returns: + torch.Tensor: The replaced input ids. + """ + input_ids_fwd = input_ids.clone() + input_ids_fwd[input_ids_fwd >= self.num_embeddings] = 0 + return input_ids_fwd + + def replace_embeddings( + self, input_ids: torch.Tensor, embedding: torch.Tensor, external_embedding: dict + ) -> torch.Tensor: + """Replace external embedding to the embedding layer. Noted that, in + this function we use `torch.cat` to avoid inplace modification. + + Args: + input_ids (torch.Tensor): The original token ids. Shape like + [LENGTH, ]. + embedding (torch.Tensor): The embedding of token ids after + `replace_input_ids` function. + external_embedding (dict): The external embedding to be replaced. + + Returns: + torch.Tensor: The replaced embedding. + """ + new_embedding = [] + + name = external_embedding["name"] + start = external_embedding["start"] + end = external_embedding["end"] + target_ids_to_replace = [i for i in range(start, end)] + ext_emb = external_embedding["embedding"] + + # do not need to replace + if not (input_ids == start).any(): + return embedding + + # start replace + s_idx, e_idx = 0, 0 + while e_idx < len(input_ids): + if input_ids[e_idx] == start: + if e_idx != 0: + # add embedding do not need to replace + new_embedding.append(embedding[s_idx:e_idx]) + + # check if the next embedding need to replace is valid + actually_ids_to_replace = [int(i) for i in input_ids[e_idx : e_idx + end - start]] + assert actually_ids_to_replace == target_ids_to_replace, ( + f"Invalid 'input_ids' in position: {s_idx} to {e_idx}. " + f"Expect '{target_ids_to_replace}' for embedding " + f"'{name}' but found '{actually_ids_to_replace}'." + ) + + new_embedding.append(ext_emb) + + s_idx = e_idx + end - start + e_idx = s_idx + 1 + else: + e_idx += 1 + + if e_idx == len(input_ids): + new_embedding.append(embedding[s_idx:e_idx]) + + return torch.cat(new_embedding, dim=0) + + def forward(self, input_ids: torch.Tensor, external_embeddings: Optional[List[dict]] = None): + """The forward function. + + Args: + input_ids (torch.Tensor): The token ids shape like [bz, LENGTH] or + [LENGTH, ]. + external_embeddings (Optional[List[dict]]): The external + embeddings. If not passed, only `self.external_embeddings` + will be used. Defaults to None. + + input_ids: shape like [bz, LENGTH] or [LENGTH]. + """ + assert input_ids.ndim in [1, 2] + if input_ids.ndim == 1: + input_ids = input_ids.unsqueeze(0) + + if external_embeddings is None and not self.external_embeddings: + return self.wrapped(input_ids) + + input_ids_fwd = self.replace_input_ids(input_ids) + inputs_embeds = self.wrapped(input_ids_fwd) + + vecs = [] + + if external_embeddings is None: + external_embeddings = [] + elif isinstance(external_embeddings, dict): + external_embeddings = [external_embeddings] + embeddings = self.external_embeddings + external_embeddings + + for input_id, embedding in zip(input_ids, inputs_embeds): + new_embedding = embedding + for external_embedding in embeddings: + new_embedding = self.replace_embeddings(input_id, new_embedding, external_embedding) + vecs.append(new_embedding) + + return torch.stack(vecs) + + + +def add_tokens( + tokenizer, text_encoder, placeholder_tokens: list, initialize_tokens: list = None, num_vectors_per_token: int = 1 +): + """Add token for training. + + # TODO: support add tokens as dict, then we can load pretrained tokens. + """ + if initialize_tokens is not None: + assert len(initialize_tokens) == len( + placeholder_tokens + ), "placeholder_token should be the same length as initialize_token" + for ii in range(len(placeholder_tokens)): + tokenizer.add_placeholder_token(placeholder_tokens[ii], num_vec_per_token=num_vectors_per_token) + + # text_encoder.set_embedding_layer() + embedding_layer = text_encoder.text_model.embeddings.token_embedding + text_encoder.text_model.embeddings.token_embedding = EmbeddingLayerWithFixes(embedding_layer) + embedding_layer = text_encoder.text_model.embeddings.token_embedding + + assert embedding_layer is not None, ( + "Do not support get embedding layer for current text encoder. " "Please check your configuration." + ) + initialize_embedding = [] + if initialize_tokens is not None: + for ii in range(len(placeholder_tokens)): + init_id = tokenizer(initialize_tokens[ii]).input_ids[1] + temp_embedding = embedding_layer.weight[init_id] + initialize_embedding.append(temp_embedding[None, ...].repeat(num_vectors_per_token, 1)) + else: + for ii in range(len(placeholder_tokens)): + init_id = tokenizer("a").input_ids[1] + temp_embedding = embedding_layer.weight[init_id] + len_emb = temp_embedding.shape[0] + init_weight = (torch.rand(num_vectors_per_token, len_emb) - 0.5) / 2.0 + initialize_embedding.append(init_weight) + + # initialize_embedding = torch.cat(initialize_embedding,dim=0) + + token_info_all = [] + for ii in range(len(placeholder_tokens)): + token_info = tokenizer.get_token_info(placeholder_tokens[ii]) + token_info["embedding"] = initialize_embedding[ii] + token_info["trainable"] = True + token_info_all.append(token_info) + embedding_layer.add_embeddings(token_info_all) diff --git a/brushnet/utils.py b/brushnet/utils.py new file mode 100644 index 0000000..75c9f5b --- /dev/null +++ b/brushnet/utils.py @@ -0,0 +1,528 @@ +import copy +import os +import random +from logging import WARNING +from typing import Any, List, Optional, Union + +import torch +import torch.nn as nn +from mmagic.utils import try_import +from mmengine import print_log + + +class TokenizerWrapper: + """Tokenizer wrapper for CLIPTokenizer. Only support CLIPTokenizer + currently. This wrapper is modified from https://github.com/huggingface/dif + fusers/blob/e51f19aee82c8dd874b715a09dbc521d88835d68/src/diffusers/loaders. + py#L358 # noqa. + + Args: + from_pretrained (Union[str, os.PathLike], optional): The *model id* + of a pretrained model or a path to a *directory* containing + model weights and config. Defaults to None. + from_config (Union[str, os.PathLike], optional): The *model id* + of a pretrained model or a path to a *directory* containing + model weights and config. Defaults to None. + + *args, **kwargs: If `from_pretrained` is passed, *args and **kwargs + will be passed to `from_pretrained` function. Otherwise, *args + and **kwargs will be used to initialize the model by + `self._module_cls(*args, **kwargs)`. + """ + + def __init__( + self, + from_pretrained: Optional[Union[str, os.PathLike]] = None, + from_config: Optional[Union[str, os.PathLike]] = None, + *args, + **kwargs, + ): + transformers = try_import("transformers") + module_cls = transformers.CLIPTokenizer + + assert not (from_pretrained and from_config), ( + "'from_pretrained' and 'from_config' should not be passed " "at the same time." + ) + + if from_config: + print_log( + "Tokenizers from Huggingface transformers do not support " + "'from_config'. Will call 'from_pretrained' instead " + "with the same argument.", + "current", + WARNING, + ) + from_pretrained = from_config + + if from_pretrained: + self.wrapped = module_cls.from_pretrained(from_pretrained, *args, **kwargs) + else: + self.wrapper = module_cls(*args, **kwargs) + + self._from_pretrained = from_pretrained + self.token_map = {} + + def __getattr__(self, name: str) -> Any: + if name == "wrapped": + return super().__getattr__("wrapped") + + try: + return getattr(self.wrapped, name) + except AttributeError: + try: + return super().__getattr__(name) + except AttributeError: + raise AttributeError( + "'name' cannot be found in both " + f"'{self.__class__.__name__}' and " + f"'{self.__class__.__name__}.tokenizer'." + ) + + def try_adding_tokens(self, tokens: Union[str, List[str]], *args, **kwargs): + """Attempt to add tokens to the tokenizer. + + Args: + tokens (Union[str, List[str]]): The tokens to be added. + """ + num_added_tokens = self.wrapped.add_tokens(tokens, *args, **kwargs) + assert num_added_tokens != 0, ( + f"The tokenizer already contains the token {tokens}. Please pass " + "a different `placeholder_token` that is not already in the " + "tokenizer." + ) + + def get_token_info(self, token: str) -> dict: + """Get the information of a token, including its start and end index in + the current tokenizer. + + Args: + token (str): The token to be queried. + + Returns: + dict: The information of the token, including its start and end + index in current tokenizer. + """ + token_ids = self.__call__(token).input_ids + start, end = token_ids[1], token_ids[-2] + 1 + return {"name": token, "start": start, "end": end} + + def add_placeholder_token(self, placeholder_token: str, *args, num_vec_per_token: int = 1, **kwargs): + """Add placeholder tokens to the tokenizer. + + Args: + placeholder_token (str): The placeholder token to be added. + num_vec_per_token (int, optional): The number of vectors of + the added placeholder token. + *args, **kwargs: The arguments for `self.wrapped.add_tokens`. + """ + output = [] + if num_vec_per_token == 1: + self.try_adding_tokens(placeholder_token, *args, **kwargs) + output.append(placeholder_token) + else: + output = [] + for i in range(num_vec_per_token): + ith_token = placeholder_token + f"_{i}" + self.try_adding_tokens(ith_token, *args, **kwargs) + output.append(ith_token) + + for token in self.token_map: + if token in placeholder_token: + raise ValueError( + f"The tokenizer already has placeholder token {token} " + f"that can get confused with {placeholder_token} " + "keep placeholder tokens independent" + ) + self.token_map[placeholder_token] = output + + def replace_placeholder_tokens_in_text( + self, text: Union[str, List[str]], vector_shuffle: bool = False, prop_tokens_to_load: float = 1.0 + ) -> Union[str, List[str]]: + """Replace the keywords in text with placeholder tokens. This function + will be called in `self.__call__` and `self.encode`. + + Args: + text (Union[str, List[str]]): The text to be processed. + vector_shuffle (bool, optional): Whether to shuffle the vectors. + Defaults to False. + prop_tokens_to_load (float, optional): The proportion of tokens to + be loaded. If 1.0, all tokens will be loaded. Defaults to 1.0. + + Returns: + Union[str, List[str]]: The processed text. + """ + if isinstance(text, list): + output = [] + for i in range(len(text)): + output.append(self.replace_placeholder_tokens_in_text(text[i], vector_shuffle=vector_shuffle)) + return output + + for placeholder_token in self.token_map: + if placeholder_token in text: + tokens = self.token_map[placeholder_token] + tokens = tokens[: 1 + int(len(tokens) * prop_tokens_to_load)] + if vector_shuffle: + tokens = copy.copy(tokens) + random.shuffle(tokens) + text = text.replace(placeholder_token, " ".join(tokens)) + return text + + def replace_text_with_placeholder_tokens(self, text: Union[str, List[str]]) -> Union[str, List[str]]: + """Replace the placeholder tokens in text with the original keywords. + This function will be called in `self.decode`. + + Args: + text (Union[str, List[str]]): The text to be processed. + + Returns: + Union[str, List[str]]: The processed text. + """ + if isinstance(text, list): + output = [] + for i in range(len(text)): + output.append(self.replace_text_with_placeholder_tokens(text[i])) + return output + + for placeholder_token, tokens in self.token_map.items(): + merged_tokens = " ".join(tokens) + if merged_tokens in text: + text = text.replace(merged_tokens, placeholder_token) + return text + + def __call__( + self, + text: Union[str, List[str]], + *args, + vector_shuffle: bool = False, + prop_tokens_to_load: float = 1.0, + **kwargs, + ): + """The call function of the wrapper. + + Args: + text (Union[str, List[str]]): The text to be tokenized. + vector_shuffle (bool, optional): Whether to shuffle the vectors. + Defaults to False. + prop_tokens_to_load (float, optional): The proportion of tokens to + be loaded. If 1.0, all tokens will be loaded. Defaults to 1.0 + *args, **kwargs: The arguments for `self.wrapped.__call__`. + """ + replaced_text = self.replace_placeholder_tokens_in_text( + text, vector_shuffle=vector_shuffle, prop_tokens_to_load=prop_tokens_to_load + ) + + return self.wrapped.__call__(replaced_text, *args, **kwargs) + + def encode(self, text: Union[str, List[str]], *args, **kwargs): + """Encode the passed text to token index. + + Args: + text (Union[str, List[str]]): The text to be encode. + *args, **kwargs: The arguments for `self.wrapped.__call__`. + """ + replaced_text = self.replace_placeholder_tokens_in_text(text) + return self.wrapped(replaced_text, *args, **kwargs) + + def decode(self, token_ids, return_raw: bool = False, *args, **kwargs) -> Union[str, List[str]]: + """Decode the token index to text. + + Args: + token_ids: The token index to be decoded. + return_raw: Whether keep the placeholder token in the text. + Defaults to False. + *args, **kwargs: The arguments for `self.wrapped.decode`. + + Returns: + Union[str, List[str]]: The decoded text. + """ + text = self.wrapped.decode(token_ids, *args, **kwargs) + if return_raw: + return text + replaced_text = self.replace_text_with_placeholder_tokens(text) + return replaced_text + + def __repr__(self): + """The representation of the wrapper.""" + s = super().__repr__() + prefix = f"Wrapped Module Class: {self._module_cls}\n" + prefix += f"Wrapped Module Name: {self._module_name}\n" + if self._from_pretrained: + prefix += f"From Pretrained: {self._from_pretrained}\n" + s = prefix + s + return s + + +class EmbeddingLayerWithFixes(nn.Module): + """The revised embedding layer to support external embeddings. This design + of this class is inspired by https://github.com/AUTOMATIC1111/stable- + diffusion-webui/blob/22bcc7be428c94e9408f589966c2040187245d81/modules/sd_hi + jack.py#L224 # noqa. + + Args: + wrapped (nn.Emebdding): The embedding layer to be wrapped. + external_embeddings (Union[dict, List[dict]], optional): The external + embeddings added to this layer. Defaults to None. + """ + + def __init__(self, wrapped: nn.Embedding, external_embeddings: Optional[Union[dict, List[dict]]] = None): + super().__init__() + self.wrapped = wrapped + self.num_embeddings = wrapped.weight.shape[0] + + self.external_embeddings = [] + if external_embeddings: + self.add_embeddings(external_embeddings) + + self.trainable_embeddings = nn.ParameterDict() + + @property + def weight(self): + """Get the weight of wrapped embedding layer.""" + return self.wrapped.weight + + def check_duplicate_names(self, embeddings: List[dict]): + """Check whether duplicate names exist in list of 'external + embeddings'. + + Args: + embeddings (List[dict]): A list of embedding to be check. + """ + names = [emb["name"] for emb in embeddings] + assert len(names) == len(set(names)), ( + "Found duplicated names in 'external_embeddings'. Name list: " f"'{names}'" + ) + + def check_ids_overlap(self, embeddings): + """Check whether overlap exist in token ids of 'external_embeddings'. + + Args: + embeddings (List[dict]): A list of embedding to be check. + """ + ids_range = [[emb["start"], emb["end"], emb["name"]] for emb in embeddings] + ids_range.sort() # sort by 'start' + # check if 'end' has overlapping + for idx in range(len(ids_range) - 1): + name1, name2 = ids_range[idx][-1], ids_range[idx + 1][-1] + assert ids_range[idx][1] <= ids_range[idx + 1][0], ( + f"Found ids overlapping between embeddings '{name1}' " f"and '{name2}'." + ) + + def add_embeddings(self, embeddings: Optional[Union[dict, List[dict]]]): + """Add external embeddings to this layer. + + Use case: + + >>> 1. Add token to tokenizer and get the token id. + >>> tokenizer = TokenizerWrapper('openai/clip-vit-base-patch32') + >>> # 'how much' in kiswahili + >>> tokenizer.add_placeholder_tokens('ngapi', num_vec_per_token=4) + >>> + >>> 2. Add external embeddings to the model. + >>> new_embedding = { + >>> 'name': 'ngapi', # 'how much' in kiswahili + >>> 'embedding': torch.ones(1, 15) * 4, + >>> 'start': tokenizer.get_token_info('kwaheri')['start'], + >>> 'end': tokenizer.get_token_info('kwaheri')['end'], + >>> 'trainable': False # if True, will registry as a parameter + >>> } + >>> embedding_layer = nn.Embedding(10, 15) + >>> embedding_layer_wrapper = EmbeddingLayerWithFixes(embedding_layer) + >>> embedding_layer_wrapper.add_embeddings(new_embedding) + >>> + >>> 3. Forward tokenizer and embedding layer! + >>> input_text = ['hello, ngapi!', 'hello my friend, ngapi?'] + >>> input_ids = tokenizer( + >>> input_text, padding='max_length', truncation=True, + >>> return_tensors='pt')['input_ids'] + >>> out_feat = embedding_layer_wrapper(input_ids) + >>> + >>> 4. Let's validate the result! + >>> assert (out_feat[0, 3: 7] == 2.3).all() + >>> assert (out_feat[2, 5: 9] == 2.3).all() + + Args: + embeddings (Union[dict, list[dict]]): The external embeddings to + be added. Each dict must contain the following 4 fields: 'name' + (the name of this embedding), 'embedding' (the embedding + tensor), 'start' (the start token id of this embedding), 'end' + (the end token id of this embedding). For example: + `{name: NAME, start: START, end: END, embedding: torch.Tensor}` + """ + if isinstance(embeddings, dict): + embeddings = [embeddings] + + self.external_embeddings += embeddings + self.check_duplicate_names(self.external_embeddings) + self.check_ids_overlap(self.external_embeddings) + + # set for trainable + added_trainable_emb_info = [] + for embedding in embeddings: + trainable = embedding.get("trainable", False) + if trainable: + name = embedding["name"] + embedding["embedding"] = torch.nn.Parameter(embedding["embedding"]) + self.trainable_embeddings[name] = embedding["embedding"] + added_trainable_emb_info.append(name) + + added_emb_info = [emb["name"] for emb in embeddings] + added_emb_info = ", ".join(added_emb_info) + print_log(f"Successfully add external embeddings: {added_emb_info}.", "current") + + if added_trainable_emb_info: + added_trainable_emb_info = ", ".join(added_trainable_emb_info) + print_log("Successfully add trainable external embeddings: " f"{added_trainable_emb_info}", "current") + + def replace_input_ids(self, input_ids: torch.Tensor) -> torch.Tensor: + """Replace external input ids to 0. + + Args: + input_ids (torch.Tensor): The input ids to be replaced. + + Returns: + torch.Tensor: The replaced input ids. + """ + input_ids_fwd = input_ids.clone() + input_ids_fwd[input_ids_fwd >= self.num_embeddings] = 0 + return input_ids_fwd + + def replace_embeddings( + self, input_ids: torch.Tensor, embedding: torch.Tensor, external_embedding: dict + ) -> torch.Tensor: + """Replace external embedding to the embedding layer. Noted that, in + this function we use `torch.cat` to avoid inplace modification. + + Args: + input_ids (torch.Tensor): The original token ids. Shape like + [LENGTH, ]. + embedding (torch.Tensor): The embedding of token ids after + `replace_input_ids` function. + external_embedding (dict): The external embedding to be replaced. + + Returns: + torch.Tensor: The replaced embedding. + """ + new_embedding = [] + + name = external_embedding["name"] + start = external_embedding["start"] + end = external_embedding["end"] + target_ids_to_replace = [i for i in range(start, end)] + ext_emb = external_embedding["embedding"] + + # do not need to replace + if not (input_ids == start).any(): + return embedding + + # start replace + s_idx, e_idx = 0, 0 + while e_idx < len(input_ids): + if input_ids[e_idx] == start: + if e_idx != 0: + # add embedding do not need to replace + new_embedding.append(embedding[s_idx:e_idx]) + + # check if the next embedding need to replace is valid + actually_ids_to_replace = [int(i) for i in input_ids[e_idx : e_idx + end - start]] + assert actually_ids_to_replace == target_ids_to_replace, ( + f"Invalid 'input_ids' in position: {s_idx} to {e_idx}. " + f"Expect '{target_ids_to_replace}' for embedding " + f"'{name}' but found '{actually_ids_to_replace}'." + ) + + new_embedding.append(ext_emb) + + s_idx = e_idx + end - start + e_idx = s_idx + 1 + else: + e_idx += 1 + + if e_idx == len(input_ids): + new_embedding.append(embedding[s_idx:e_idx]) + + return torch.cat(new_embedding, dim=0) + + def forward(self, input_ids: torch.Tensor, external_embeddings: Optional[List[dict]] = None): + """The forward function. + + Args: + input_ids (torch.Tensor): The token ids shape like [bz, LENGTH] or + [LENGTH, ]. + external_embeddings (Optional[List[dict]]): The external + embeddings. If not passed, only `self.external_embeddings` + will be used. Defaults to None. + + input_ids: shape like [bz, LENGTH] or [LENGTH]. + """ + assert input_ids.ndim in [1, 2] + if input_ids.ndim == 1: + input_ids = input_ids.unsqueeze(0) + + if external_embeddings is None and not self.external_embeddings: + return self.wrapped(input_ids) + + input_ids_fwd = self.replace_input_ids(input_ids) + inputs_embeds = self.wrapped(input_ids_fwd) + + vecs = [] + + if external_embeddings is None: + external_embeddings = [] + elif isinstance(external_embeddings, dict): + external_embeddings = [external_embeddings] + embeddings = self.external_embeddings + external_embeddings + + for input_id, embedding in zip(input_ids, inputs_embeds): + new_embedding = embedding + for external_embedding in embeddings: + new_embedding = self.replace_embeddings(input_id, new_embedding, external_embedding) + vecs.append(new_embedding) + + return torch.stack(vecs) + + +def add_tokens( + tokenizer, text_encoder, placeholder_tokens: list, initialize_tokens: list = None, num_vectors_per_token: int = 1 +): + """Add token for training. + + # TODO: support add tokens as dict, then we can load pretrained tokens. + """ + if initialize_tokens is not None: + assert len(initialize_tokens) == len( + placeholder_tokens + ), "placeholder_token should be the same length as initialize_token" + for ii in range(len(placeholder_tokens)): + tokenizer.add_placeholder_token(placeholder_tokens[ii], num_vec_per_token=num_vectors_per_token) + + # text_encoder.set_embedding_layer() + embedding_layer = text_encoder.text_model.embeddings.token_embedding + text_encoder.text_model.embeddings.token_embedding = EmbeddingLayerWithFixes(embedding_layer) + embedding_layer = text_encoder.text_model.embeddings.token_embedding + + assert embedding_layer is not None, ( + "Do not support get embedding layer for current text encoder. " "Please check your configuration." + ) + initialize_embedding = [] + if initialize_tokens is not None: + for ii in range(len(placeholder_tokens)): + init_id = tokenizer(initialize_tokens[ii]).input_ids[1] + temp_embedding = embedding_layer.weight[init_id] + initialize_embedding.append(temp_embedding[None, ...].repeat(num_vectors_per_token, 1)) + else: + for ii in range(len(placeholder_tokens)): + init_id = tokenizer("a").input_ids[1] + temp_embedding = embedding_layer.weight[init_id] + len_emb = temp_embedding.shape[0] + init_weight = (torch.rand(num_vectors_per_token, len_emb) - 0.5) / 2.0 + initialize_embedding.append(init_weight) + + # initialize_embedding = torch.cat(initialize_embedding,dim=0) + + token_info_all = [] + for ii in range(len(placeholder_tokens)): + token_info = tokenizer.get_token_info(placeholder_tokens[ii]) + token_info["embedding"] = initialize_embedding[ii] + token_info["trainable"] = True + token_info_all.append(token_info) + embedding_layer.add_embeddings(token_info_all) diff --git a/brushnet_nodes.py b/brushnet_nodes.py index d5f0e6e..14cb0c6 100644 --- a/brushnet_nodes.py +++ b/brushnet_nodes.py @@ -15,6 +15,7 @@ import latent_preview from accelerate import init_empty_weights, load_checkpoint_and_dispatch from .brushnet.brushnet import BrushNetModel +from .brushnet.brushnet_ca import BrushNetModel as PowerPaintModel import types from typing import Tuple @@ -25,6 +26,7 @@ warnings.filterwarnings("ignore", category=UserWarning, module="safetensors") current_directory = os.path.dirname(os.path.abspath(__file__)) brushnet_config_file = os.path.join(current_directory, 'brushnet', 'brushnet.json') brushnet_xl_config_file = os.path.join(current_directory, 'brushnet', 'brushnet_xl.json') +powerpaint_config_file = os.path.join(current_directory,'brushnet', 'powerpaint.json') torch_dtype = torch.float16 sd15_scaling_factor = 0.18215 @@ -50,24 +52,35 @@ class BrushNetLoader: brushnet_file = os.path.join(folder_paths.models_dir, "inpaint", brushnet) is_SDXL = False + is_PP = False sd = comfy.utils.load_torch_file(brushnet_file) - brushnet_down_block, brushnet_mid_block, brushnet_up_block = brushnet_blocks(sd) + brushnet_down_block, brushnet_mid_block, brushnet_up_block, keys = brushnet_blocks(sd) del sd if brushnet_down_block == 24 and brushnet_mid_block == 2 and brushnet_up_block == 30: - print('BrushNet model type: SD1.5') is_SDXL = False + if keys == 322: + is_PP = False + print('BrushNet model type: SD1.5') + else: + is_PP = True + print('PowerPaint model type: SD1.5') elif brushnet_down_block == 18 and brushnet_mid_block == 2 and brushnet_up_block == 22: print('BrushNet model type: Loading SDXL') is_SDXL = True + is_PP = False else: raise Exception("Unknown BrushNet model") with init_empty_weights(): if is_SDXL: brushnet_config = BrushNetModel.load_config(brushnet_xl_config_file) + brushnet_model = BrushNetModel.from_config(brushnet_config) + elif is_PP: + brushnet_config = PowerPaintModel.load_config(powerpaint_config_file) + brushnet_model = PowerPaintModel.from_config(brushnet_config) else: brushnet_config = BrushNetModel.load_config(brushnet_config_file) - brushnet_model = BrushNetModel.from_config(brushnet_config) + brushnet_model = BrushNetModel.from_config(brushnet_config) print("BrushNet model file:", brushnet_file) @@ -82,9 +95,12 @@ class BrushNetLoader: force_hooks=False, ) - print("BrushNet model is loaded, SDXL:", is_SDXL) + if is_PP: + print("PowerPaint model is loaded") + else: + print("BrushNet model is loaded, SDXL:", is_SDXL) - return ({"brushnet": brushnet_model, "SDXL": is_SDXL},) + return ({"brushnet": brushnet_model, "SDXL": is_SDXL, "PP": is_PP},) def inpaint_safetensors(): @@ -119,7 +135,7 @@ def brushnet_blocks(sd): brushnet_mid_block += 1 if 'brushnet_up_block' in key: brushnet_up_block += 1 - return (brushnet_down_block, brushnet_mid_block, brushnet_up_block, ) + return (brushnet_down_block, brushnet_mid_block, brushnet_up_block, len(sd)) class BrushNet: @@ -139,11 +155,15 @@ class BrushNet: "start_at": ("FLOAT", {"default": 0.0, "min": 0.0, "max": 1.0}), "end_at": ("FLOAT", {"default": 1.0, "min": 0.0, "max": 1.0}), }, - } + "optional": + { + "clip": ("PPCLIP",), + }, + } CATEGORY = "inpaint" - RETURN_TYPES = ("MODEL","LATENT",) - RETURN_NAMES = ("model","latent",) + RETURN_TYPES = ("MODEL","CONDITIONING","CONDITIONING","LATENT",) + RETURN_NAMES = ("model","positive","negative","latent",) FUNCTION = "model_update" @@ -154,7 +174,9 @@ class BrushNet: print('Base model type: SD1.5') is_SDXL = False if brushnet["SDXL"]: - raise Exception("Base model is SD15, but BrushNet is SDXL type") + raise Exception("Base model is SD15, but BrushNet is SDXL type") + if brushnet["PP"]: + raise Exception("PowerPaint model is not yet supported") elif isinstance(model.model.model_config, comfy.supported_models.SDXL): print('Base model type: SDXL') is_SDXL = True @@ -263,7 +285,7 @@ class BrushNet: latent = torch.zeros([1, 4, conditioning_latents.shape[2], conditioning_latents.shape[3]], device=brushnet['brushnet'].device) #latent = torch.randn([1, 4, conditioning_latents.shape[2], conditioning_latents.shape[3]], device=brushnet['brushnet'].device) - return (cloned, {"samples":latent},) + return (cloned, positive, negative, {"samples":latent},) class BlendInpaint: