Initial commit

.gitignore

@@ -0,0 +1,6 @@
+__pycache__/
+checkpoints/
+*.py[cod]
+*$py.class
+*.egg-info
+.pytest_cache

LICENSE

@@ -1,4 +1,4 @@
-Apache License
+                                 Apache License
                            Version 2.0, January 2004
                         http://www.apache.org/licenses/
 

README.md

@@ -0,0 +1,16 @@
+# ComfyUI wrapper nodes to use the Diffusers implementation of BrushNet
+
+
+
+# Installing
+Either use the Manager and it's install from git -feature, or clone this repo to custom_nodes and run:
+
+`pip install -r requirements.txt`
+
+or if you use portable (run this in ComfyUI_windows_portable -folder):
+
+`python_embeded\python.exe -m pip install -r ComfyUI\custom_nodes\ComfyUI-BrushNetWrapper\requirements.txt`
+
+Models are automatically downloaded from here: https://huggingface.co/Kijai/BrushNet-fp16/tree/main
+
+to `ComfyUI/models/brushnet`

__init__.py

@@ -0,0 +1,3 @@
+from .nodes import NODE_CLASS_MAPPINGS, NODE_DISPLAY_NAME_MAPPINGS
+
+__all__ = ["NODE_CLASS_MAPPINGS", "NODE_DISPLAY_NAME_MAPPINGS"]

brushnet/.gitignore

@@ -0,0 +1,6 @@
+__pycache__/
+checkpoints/
+*.py[cod]
+*$py.class
+*.egg-info
+.pytest_cache

configs/brushnet_config.json

@@ -0,0 +1,58 @@
+{
+  "_class_name": "BrushNetModel",
+  "_diffusers_version": "0.27.0.dev0",
+  "_name_or_path": "runs/logs/brushnet_segmask/checkpoint-550000",
+  "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": [
+    "DownBlock2D",
+    "DownBlock2D",
+    "DownBlock2D",
+    "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": "MidBlock2D",
+  "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",
+    "UpBlock2D",
+    "UpBlock2D",
+    "UpBlock2D"
+  ],
+  "upcast_attention": false,
+  "use_linear_projection": false
+}

configs/text_encoder_config.json

@@ -0,0 +1,25 @@
+{
+  "_name_or_path": "openai/clip-vit-large-patch14",
+  "architectures": [
+    "CLIPTextModel"
+  ],
+  "attention_dropout": 0.0,
+  "bos_token_id": 0,
+  "dropout": 0.0,
+  "eos_token_id": 2,
+  "hidden_act": "quick_gelu",
+  "hidden_size": 768,
+  "initializer_factor": 1.0,
+  "initializer_range": 0.02,
+  "intermediate_size": 3072,
+  "layer_norm_eps": 1e-05,
+  "max_position_embeddings": 77,
+  "model_type": "clip_text_model",
+  "num_attention_heads": 12,
+  "num_hidden_layers": 12,
+  "pad_token_id": 1,
+  "projection_dim": 768,
+  "torch_dtype": "float32",
+  "transformers_version": "4.22.0.dev0",
+  "vocab_size": 49408
+}

configs/tokenizer/config.json

@@ -0,0 +1,171 @@
+{
+  "_name_or_path": "clip-vit-large-patch14/",
+  "architectures": [
+    "CLIPModel"
+  ],
+  "initializer_factor": 1.0,
+  "logit_scale_init_value": 2.6592,
+  "model_type": "clip",
+  "projection_dim": 768,
+  "text_config": {
+    "_name_or_path": "",
+    "add_cross_attention": false,
+    "architectures": null,
+    "attention_dropout": 0.0,
+    "bad_words_ids": null,
+    "bos_token_id": 0,
+    "chunk_size_feed_forward": 0,
+    "cross_attention_hidden_size": null,
+    "decoder_start_token_id": null,
+    "diversity_penalty": 0.0,
+    "do_sample": false,
+    "dropout": 0.0,
+    "early_stopping": false,
+    "encoder_no_repeat_ngram_size": 0,
+    "eos_token_id": 2,
+    "finetuning_task": null,
+    "forced_bos_token_id": null,
+    "forced_eos_token_id": null,
+    "hidden_act": "quick_gelu",
+    "hidden_size": 768,
+    "id2label": {
+      "0": "LABEL_0",
+      "1": "LABEL_1"
+    },
+    "initializer_factor": 1.0,
+    "initializer_range": 0.02,
+    "intermediate_size": 3072,
+    "is_decoder": false,
+    "is_encoder_decoder": false,
+    "label2id": {
+      "LABEL_0": 0,
+      "LABEL_1": 1
+    },
+    "layer_norm_eps": 1e-05,
+    "length_penalty": 1.0,
+    "max_length": 20,
+    "max_position_embeddings": 77,
+    "min_length": 0,
+    "model_type": "clip_text_model",
+    "no_repeat_ngram_size": 0,
+    "num_attention_heads": 12,
+    "num_beam_groups": 1,
+    "num_beams": 1,
+    "num_hidden_layers": 12,
+    "num_return_sequences": 1,
+    "output_attentions": false,
+    "output_hidden_states": false,
+    "output_scores": false,
+    "pad_token_id": 1,
+    "prefix": null,
+    "problem_type": null,
+    "projection_dim" : 768,
+    "pruned_heads": {},
+    "remove_invalid_values": false,
+    "repetition_penalty": 1.0,
+    "return_dict": true,
+    "return_dict_in_generate": false,
+    "sep_token_id": null,
+    "task_specific_params": null,
+    "temperature": 1.0,
+    "tie_encoder_decoder": false,
+    "tie_word_embeddings": true,
+    "tokenizer_class": null,
+    "top_k": 50,
+    "top_p": 1.0,
+    "torch_dtype": null,
+    "torchscript": false,
+    "transformers_version": "4.16.0.dev0",
+    "use_bfloat16": false,
+    "vocab_size": 49408
+  },
+  "text_config_dict": {
+    "hidden_size": 768,
+    "intermediate_size": 3072,
+    "num_attention_heads": 12,
+    "num_hidden_layers": 12,
+    "projection_dim": 768
+  },
+  "torch_dtype": "float32",
+  "transformers_version": null,
+  "vision_config": {
+    "_name_or_path": "",
+    "add_cross_attention": false,
+    "architectures": null,
+    "attention_dropout": 0.0,
+    "bad_words_ids": null,
+    "bos_token_id": null,
+    "chunk_size_feed_forward": 0,
+    "cross_attention_hidden_size": null,
+    "decoder_start_token_id": null,
+    "diversity_penalty": 0.0,
+    "do_sample": false,
+    "dropout": 0.0,
+    "early_stopping": false,
+    "encoder_no_repeat_ngram_size": 0,
+    "eos_token_id": null,
+    "finetuning_task": null,
+    "forced_bos_token_id": null,
+    "forced_eos_token_id": null,
+    "hidden_act": "quick_gelu",
+    "hidden_size": 1024,
+    "id2label": {
+      "0": "LABEL_0",
+      "1": "LABEL_1"
+    },
+    "image_size": 224,
+    "initializer_factor": 1.0,
+    "initializer_range": 0.02,
+    "intermediate_size": 4096,
+    "is_decoder": false,
+    "is_encoder_decoder": false,
+    "label2id": {
+      "LABEL_0": 0,
+      "LABEL_1": 1
+    },
+    "layer_norm_eps": 1e-05,
+    "length_penalty": 1.0,
+    "max_length": 20,
+    "min_length": 0,
+    "model_type": "clip_vision_model",
+    "no_repeat_ngram_size": 0,
+    "num_attention_heads": 16,
+    "num_beam_groups": 1,
+    "num_beams": 1,
+    "num_hidden_layers": 24,
+    "num_return_sequences": 1,
+    "output_attentions": false,
+    "output_hidden_states": false,
+    "output_scores": false,
+    "pad_token_id": null,
+    "patch_size": 14,
+    "prefix": null,
+    "problem_type": null,
+    "projection_dim" : 768,
+    "pruned_heads": {},
+    "remove_invalid_values": false,
+    "repetition_penalty": 1.0,
+    "return_dict": true,
+    "return_dict_in_generate": false,
+    "sep_token_id": null,
+    "task_specific_params": null,
+    "temperature": 1.0,
+    "tie_encoder_decoder": false,
+    "tie_word_embeddings": true,
+    "tokenizer_class": null,
+    "top_k": 50,
+    "top_p": 1.0,
+    "torch_dtype": null,
+    "torchscript": false,
+    "transformers_version": "4.16.0.dev0",
+    "use_bfloat16": false
+  },
+  "vision_config_dict": {
+    "hidden_size": 1024,
+    "intermediate_size": 4096,
+    "num_attention_heads": 16,
+    "num_hidden_layers": 24,
+    "patch_size": 14,
+    "projection_dim": 768
+  }
+}

configs/tokenizer/preprocessor_config.json

@@ -0,0 +1,19 @@
+{
+  "crop_size": 224,
+  "do_center_crop": true,
+  "do_normalize": true,
+  "do_resize": true,
+  "feature_extractor_type": "CLIPFeatureExtractor",
+  "image_mean": [
+    0.48145466,
+    0.4578275,
+    0.40821073
+  ],
+  "image_std": [
+    0.26862954,
+    0.26130258,
+    0.27577711
+  ],
+  "resample": 3,
+  "size": 224
+}

configs/tokenizer/special_tokens_map.json

@@ -0,0 +1 @@
+{"bos_token": {"content": "<|startoftext|>", "single_word": false, "lstrip": false, "rstrip": false, "normalized": true}, "eos_token": {"content": "<|endoftext|>", "single_word": false, "lstrip": false, "rstrip": false, "normalized": true}, "unk_token": {"content": "<|endoftext|>", "single_word": false, "lstrip": false, "rstrip": false, "normalized": true}, "pad_token": "<|endoftext|>"}

configs/tokenizer/tokenizer_config.json

@@ -0,0 +1,34 @@
+{
+    "unk_token": {
+        "content": "<|endoftext|>",
+        "single_word": false,
+        "lstrip": false,
+        "rstrip": false,
+        "normalized": true,
+        "__type": "AddedToken"
+    },
+    "bos_token": {
+        "content": "<|startoftext|>",
+        "single_word": false,
+        "lstrip": false,
+        "rstrip": false,
+        "normalized": true,
+        "__type": "AddedToken"
+    },
+    "eos_token": {
+        "content": "<|endoftext|>",
+        "single_word": false,
+        "lstrip": false,
+        "rstrip": false,
+        "normalized": true,
+        "__type": "AddedToken"
+    },
+    "pad_token": "<|endoftext|>",
+    "add_prefix_space": false,
+    "errors": "replace",
+    "do_lower_case": true,
+    "name_or_path": "openai/clip-vit-base-patch32",
+    "model_max_length": 77,
+    "special_tokens_map_file": "./special_tokens_map.json",
+    "tokenizer_class": "CLIPTokenizer"
+}

configs/tokenizer_config.json

@@ -0,0 +1,34 @@
+{
+  "add_prefix_space": false,
+  "bos_token": {
+    "__type": "AddedToken",
+    "content": "<|startoftext|>",
+    "lstrip": false,
+    "normalized": true,
+    "rstrip": false,
+    "single_word": false
+  },
+  "do_lower_case": true,
+  "eos_token": {
+    "__type": "AddedToken",
+    "content": "<|endoftext|>",
+    "lstrip": false,
+    "normalized": true,
+    "rstrip": false,
+    "single_word": false
+  },
+  "errors": "replace",
+  "model_max_length": 77,
+  "name_or_path": "openai/clip-vit-large-patch14",
+  "pad_token": "<|endoftext|>",
+  "special_tokens_map_file": "./special_tokens_map.json",
+  "tokenizer_class": "CLIPTokenizer",
+  "unk_token": {
+    "__type": "AddedToken",
+    "content": "<|endoftext|>",
+    "lstrip": false,
+    "normalized": true,
+    "rstrip": false,
+    "single_word": false
+  }
+}

configs/v1-inference.yaml

@@ -0,0 +1,70 @@
+model:
+  base_learning_rate: 1.0e-04
+  target: ldm.models.diffusion.ddpm.LatentDiffusion
+  params:
+    linear_start: 0.00085
+    linear_end: 0.0120
+    num_timesteps_cond: 1
+    log_every_t: 200
+    timesteps: 1000
+    first_stage_key: "jpg"
+    cond_stage_key: "txt"
+    image_size: 64
+    channels: 4
+    cond_stage_trainable: false   # Note: different from the one we trained before
+    conditioning_key: crossattn
+    monitor: val/loss_simple_ema
+    scale_factor: 0.18215
+    use_ema: False
+
+    scheduler_config: # 10000 warmup steps
+      target: ldm.lr_scheduler.LambdaLinearScheduler
+      params:
+        warm_up_steps: [ 10000 ]
+        cycle_lengths: [ 10000000000000 ] # incredibly large number to prevent corner cases
+        f_start: [ 1.e-6 ]
+        f_max: [ 1. ]
+        f_min: [ 1. ]
+
+    unet_config:
+      target: ldm.modules.diffusionmodules.openaimodel.UNetModel
+      params:
+        image_size: 32 # unused
+        in_channels: 4
+        out_channels: 4
+        model_channels: 320
+        attention_resolutions: [ 4, 2, 1 ]
+        num_res_blocks: 2
+        channel_mult: [ 1, 2, 4, 4 ]
+        num_heads: 8
+        use_spatial_transformer: True
+        transformer_depth: 1
+        context_dim: 768
+        use_checkpoint: True
+        legacy: False
+
+    first_stage_config:
+      target: ldm.models.autoencoder.AutoencoderKL
+      params:
+        embed_dim: 4
+        monitor: val/rec_loss
+        ddconfig:
+          double_z: true
+          z_channels: 4
+          resolution: 256
+          in_channels: 3
+          out_ch: 3
+          ch: 128
+          ch_mult:
+          - 1
+          - 2
+          - 4
+          - 4
+          num_res_blocks: 2
+          attn_resolutions: []
+          dropout: 0.0
+        lossconfig:
+          target: torch.nn.Identity
+
+    cond_stage_config:
+      target: ldm.modules.encoders.modules.FrozenCLIPEmbedder

inference.py

@@ -0,0 +1,419 @@
+from pathlib import Path
+from typing import Any, Optional, Union
+
+import fire
+import gradio as gr
+import safetensors.torch
+import torch
+from diffusers import DPMSolverMultistepScheduler, StableDiffusionPipeline
+from torchvision.utils import save_image
+
+from model import ELLA, T5TextEmbedder
+
+
+class ELLAProxyUNet(torch.nn.Module):
+    def __init__(self, ella, unet):
+        super().__init__()
+        # In order to still use the diffusers pipeline, including various workaround
+
+        self.ella = ella
+        self.unet = unet
+        self.config = unet.config
+        self.dtype = unet.dtype
+        self.device = unet.device
+
+        self.flexible_max_length_workaround = None
+
+    def forward(
+        self,
+        sample: torch.FloatTensor,
+        timestep: Union[torch.Tensor, float, int],
+        encoder_hidden_states: torch.Tensor,
+        class_labels: Optional[torch.Tensor] = None,
+        timestep_cond: Optional[torch.Tensor] = None,
+        attention_mask: Optional[torch.Tensor] = None,
+        cross_attention_kwargs: Optional[dict[str, Any]] = None,
+        added_cond_kwargs: Optional[dict[str, torch.Tensor]] = None,
+        down_block_additional_residuals: Optional[tuple[torch.Tensor]] = None,
+        mid_block_additional_residual: Optional[torch.Tensor] = None,
+        down_intrablock_additional_residuals: Optional[tuple[torch.Tensor]] = None,
+        encoder_attention_mask: Optional[torch.Tensor] = None,
+        return_dict: bool = True,
+    ):
+        if self.flexible_max_length_workaround is not None:
+            time_aware_encoder_hidden_state_list = []
+            for i, max_length in enumerate(self.flexible_max_length_workaround):
+                time_aware_encoder_hidden_state_list.append(
+                    self.ella(encoder_hidden_states[i : i + 1, :max_length], timestep)
+                )
+            # No matter how many tokens are text features, the ella output must be 64 tokens.
+            time_aware_encoder_hidden_states = torch.cat(
+                time_aware_encoder_hidden_state_list, dim=0
+            )
+        else:
+            time_aware_encoder_hidden_states = self.ella(
+                encoder_hidden_states, timestep
+            )
+
+        return self.unet(
+            sample=sample,
+            timestep=timestep,
+            encoder_hidden_states=time_aware_encoder_hidden_states,
+            class_labels=class_labels,
+            timestep_cond=timestep_cond,
+            attention_mask=attention_mask,
+            cross_attention_kwargs=cross_attention_kwargs,
+            added_cond_kwargs=added_cond_kwargs,
+            down_block_additional_residuals=down_block_additional_residuals,
+            mid_block_additional_residual=mid_block_additional_residual,
+            down_intrablock_additional_residuals=down_intrablock_additional_residuals,
+            encoder_attention_mask=encoder_attention_mask,
+            return_dict=return_dict,
+        )
+
+
+def generate_image_with_flexible_max_length(
+    pipe, t5_encoder, prompt, fixed_negative=False, output_type="pt", **pipe_kwargs
+):
+    device = pipe.device
+    dtype = pipe.dtype
+    prompt = [prompt] if isinstance(prompt, str) else prompt
+    batch_size = len(prompt)
+
+    prompt_embeds = t5_encoder(prompt, max_length=None).to(device, dtype)
+    negative_prompt_embeds = t5_encoder(
+        [""] * batch_size, max_length=128 if fixed_negative else None
+    ).to(device, dtype)
+
+    # diffusers pipeline concatenate `prompt_embeds` too early...
+    # https://github.com/huggingface/diffusers/blob/b6d7e31d10df675d86c6fe7838044712c6dca4e9/src/diffusers/pipelines/stable_diffusion/pipeline_stable_diffusion.py#L913
+    pipe.unet.flexible_max_length_workaround = [
+        negative_prompt_embeds.size(1)
+    ] * batch_size + [prompt_embeds.size(1)] * batch_size
+
+    max_length = max([prompt_embeds.size(1), negative_prompt_embeds.size(1)])
+    b, _, d = prompt_embeds.shape
+    prompt_embeds = torch.cat(
+        [
+            prompt_embeds,
+            torch.zeros(
+                (b, max_length - prompt_embeds.size(1), d), device=device, dtype=dtype
+            ),
+        ],
+        dim=1,
+    )
+    negative_prompt_embeds = torch.cat(
+        [
+            negative_prompt_embeds,
+            torch.zeros(
+                (b, max_length - negative_prompt_embeds.size(1), d),
+                device=device,
+                dtype=dtype,
+            ),
+        ],
+        dim=1,
+    )
+
+    images = pipe(
+        prompt_embeds=prompt_embeds,
+        negative_prompt_embeds=negative_prompt_embeds,
+        **pipe_kwargs,
+        output_type=output_type,
+    ).images
+    pipe.unet.flexible_max_length_workaround = None
+    return images
+
+
+def load_ella(filename, device, dtype):
+    ella = ELLA()
+    safetensors.torch.load_model(ella, filename, strict=True)
+    ella.to(device, dtype=dtype)
+    return ella
+
+
+def load_ella_for_pipe(pipe, ella):
+    pipe.unet = ELLAProxyUNet(ella, pipe.unet)
+
+
+def offload_ella_for_pipe(pipe):
+    pipe.unet = pipe.unet.unet
+
+
+def generate_image_with_fixed_max_length(
+    pipe, t5_encoder, prompt, output_type="pt", **pipe_kwargs
+):
+    prompt = [prompt] if isinstance(prompt, str) else prompt
+
+    prompt_embeds = t5_encoder(prompt, max_length=128).to(pipe.device, pipe.dtype)
+    negative_prompt_embeds = t5_encoder([""] * len(prompt), max_length=128).to(
+        pipe.device, pipe.dtype
+    )
+
+    return pipe(
+        prompt_embeds=prompt_embeds,
+        negative_prompt_embeds=negative_prompt_embeds,
+        **pipe_kwargs,
+        output_type=output_type,
+    ).images
+
+
+def build_demo(ella_path, sd_path="runwayml/stable-diffusion-v1-5"):
+    pipe = StableDiffusionPipeline.from_pretrained(
+        sd_path,
+        torch_dtype=torch.float16,
+        safety_checker=None,
+        feature_extractor=None,
+        requires_safety_checker=False,
+    )
+    pipe = pipe.to("cuda")
+    pipe.scheduler = DPMSolverMultistepScheduler.from_config(pipe.scheduler.config)
+
+    ella = load_ella(ella_path, pipe.device, pipe.dtype)
+    t5_encoder = T5TextEmbedder().to(pipe.device, dtype=torch.float16)
+
+    def generate_images(
+        prompt, guidance_scale, seed, num_inference_steps, size=512, _batch_size=2
+    ):
+        print("#" * 50)
+        print(prompt)
+        load_ella_for_pipe(pipe, ella)
+        image_flexible = generate_image_with_flexible_max_length(
+            pipe,
+            t5_encoder,
+            [prompt] * _batch_size,
+            guidance_scale=guidance_scale,
+            num_inference_steps=num_inference_steps,
+            height=size,
+            width=size,
+            generator=[
+                torch.Generator(device="cuda").manual_seed(seed + i)
+                for i in range(_batch_size)
+            ],
+            output_type="pil",
+        )
+        offload_ella_for_pipe(pipe)
+
+        image_ori = pipe(
+            [prompt] * _batch_size,
+            output_type="pil",
+            guidance_scale=guidance_scale,
+            num_inference_steps=num_inference_steps,
+            height=size,
+            width=size,
+            generator=[
+                torch.Generator(device="cuda").manual_seed(seed + i)
+                for i in range(_batch_size)
+            ],
+        ).images
+
+        return image_ori, image_flexible
+
+    with gr.Blocks() as app:
+        gr.Markdown(
+            """
+    # ELLA-SD1.5 vs SD1.5
+
+    [ELLA Project](https://ella-diffusion.github.io/)
+
+    ## Notes
+    
+    ** short prompt also works, but the result is much better after the caption is refined. **
+
+    ### Caption Refining with In Context Learning(ICL)
+
+    caption refining instruction example:
+    ```
+    Please generate the long prompt version of the short one according to the given examples. Long prompt version should consist of 3 to 5 sentences. Long prompt version must sepcify the color, shape, texture or spatial relation of the included objects. DO NOT generate sentences that describe any atmosphere!!!
+
+    Short: A calico cat with eyes closed is perched upon a Mercedes.
+    Long: a multicolored cat perched atop a shiny black car. the car is parked in front of a building with wooden walls and a green fence. the reflection of the car and the surrounding environment can be seen on the car's glossy surface.
+
+    Short: A boys sitting on a chair holding a video game remote.
+    Long: a young boy sitting on a chair, wearing a blue shirt and a baseball cap with the letter 'm'. he has a red medal around his neck and is holding a white game controller. behind him, there are two other individuals, one of whom is wearing a backpack. to the right of the boy, there's a blue trash bin with a sign that reads 'automatic party'.
+
+    Short: A man is on the bank of the water fishing.
+    Long: a serene waterscape where a person, dressed in a blue jacket and a red beanie, stands in shallow waters, fishing with a long rod. the calm waters are dotted with several sailboats anchored at a distance, and a mountain range can be seen in the background under a cloudy sky.
+
+    Short: A kitchen with a cluttered counter and wooden cabinets.
+    Long: a well-lit kitchen with wooden cabinets, a black and white checkered floor, and a refrigerator adorned with a floral decal on its side. the kitchen countertop holds various items, including a coffee maker, jars, and fruits.
+
+    Short: a racoon holding a shiny red apple over its head
+    ```
+
+    using: https://huggingface.co/spaces/Qwen/Qwen-72B-Chat-Demo
+    got: a mischievous raccoon standing on its hind legs, holding a bright red apple aloft in its furry paws. the apple shines brightly against the backdrop of a dense forest, with leaves rustling in the gentle breeze. a few scattered rocks can be seen on the ground beneath the raccoon's feet, while a gnarled tree trunk stands nearby.
+            """
+        )
+        with gr.Row():
+            input_caption = gr.Textbox(
+                value="A vivid red book with a smooth, matte cover lies next to a glossy yellow vase. The vase, with a slightly curved silhouette, stands on a dark wood table with a noticeable grain pattern. The book appears slightly worn at the edges, suggesting frequent use, while the vase holds a fresh array of multicolored wildflowers."
+            )
+            with gr.Column():
+                guidance_scale = gr.Slider(
+                    minimum=1.0, maximum=16.0, value=10, label="guidance_scale"
+                )
+                seed = gr.Slider(
+                    minimum=1000, maximum=2**20, value=1000, label="random seed"
+                )
+                num_inference_steps = gr.Slider(
+                    minimum=15, maximum=100, value=25, label="num_inference_steps"
+                )
+
+        with gr.Row():
+            with gr.Column():
+                gr.Markdown(f"### ORIGINAL Stable Diffusion Model")
+                sd_output_image_gallery = gr.Gallery(columns=2, label="ORIGINAL SD")
+            with gr.Column():
+                gr.Markdown(f"### ELLA")
+                ella_output_image_gallery = gr.Gallery(columns=2, label="ELLA")
+        submit_button = gr.Button()
+
+        submit_button.click(
+            fn=generate_images,
+            inputs=[input_caption, guidance_scale, seed, num_inference_steps],
+            outputs=[sd_output_image_gallery, ella_output_image_gallery],
+        )
+
+    app.queue(concurrency_count=1, api_open=False)
+    app.launch(share=False)
+
+
+def main(save_folder, ella_path):
+    save_folder = Path(save_folder)
+    save_folder.mkdir(exist_ok=True)
+
+    pipe = StableDiffusionPipeline.from_pretrained(
+        "runwayml/stable-diffusion-v1-5",
+        torch_dtype=torch.float16,
+        safety_checker=None,
+        feature_extractor=None,
+        requires_safety_checker=False,
+    )
+    pipe = pipe.to("cuda")
+    pipe.scheduler = DPMSolverMultistepScheduler.from_config(pipe.scheduler.config)
+
+    ella = load_ella(ella_path, pipe.device, pipe.dtype)
+    t5_encoder = T5TextEmbedder().to(pipe.device, dtype=torch.float16)
+
+    # prompt from ViLG-300, PartiPrompts
+    # short prompt also works, but the result is much better after the caption is refined.
+
+    # caption refining instruction example:
+    # ```
+    # Please generate the long prompt version of the short one according to the given examples. Long prompt version should consist of 3 to 5 sentences. Long prompt version must sepcify the color, shape, texture or spatial relation of the included objects. DO NOT generate sentences that describe any atmosphere!!!
+    #
+    # Short: A calico cat with eyes closed is perched upon a Mercedes.
+    # Long: a multicolored cat perched atop a shiny black car. the car is parked in front of a building with wooden walls and a green fence. the reflection of the car and the surrounding environment can be seen on the car's glossy surface.
+    #
+    # Short: A boys sitting on a chair holding a video game remote.
+    # Long: a young boy sitting on a chair, wearing a blue shirt and a baseball cap with the letter 'm'. he has a red medal around his neck and is holding a white game controller. behind him, there are two other individuals, one of whom is wearing a backpack. to the right of the boy, there's a blue trash bin with a sign that reads 'automatic party'.
+    #
+    # Short: A man is on the bank of the water fishing.
+    # Long: a serene waterscape where a person, dressed in a blue jacket and a red beanie, stands in shallow waters, fishing with a long rod. the calm waters are dotted with several sailboats anchored at a distance, and a mountain range can be seen in the background under a cloudy sky.
+    #
+    # Short: A kitchen with a cluttered counter and wooden cabinets.
+    # Long: a well-lit kitchen with wooden cabinets, a black and white checkered floor, and a refrigerator adorned with a floral decal on its side. the kitchen countertop holds various items, including a coffee maker, jars, and fruits.
+    #
+    # Short: a racoon holding a shiny red apple over its head
+    # ```
+    #
+    # using: https://huggingface.co/spaces/Qwen/Qwen-72B-Chat-Demo
+    # got: a mischievous raccoon standing on its hind legs, holding a bright red apple aloft in its furry paws. the apple shines brightly against the backdrop of a dense forest, with leaves rustling in the gentle breeze. a few scattered rocks can be seen on the ground beneath the raccoon's feet, while a gnarled tree trunk stands nearby.
+
+    prompt_name_examples1 = [
+        ("crocodile_sweater", "Crocodile in a sweater"),
+        (
+            "crocodile_sweater-gpt4_refined_caption",
+            "a large, textured green crocodile lying comfortably on a patch of grass with a cute, knitted orange sweater enveloping its scaly body. Around its neck, the sweater features a whimsical pattern of blue and yellow stripes. In the background, a smooth, grey rock partially obscures the view of a small pond with lily pads floating on the surface.",
+        ),
+        ("red_book-yellow_vase", "A red book and a yellow vase."),
+        (
+            "red_book-yellow_vase-gpt4_refined_caption",
+            "A vivid red book with a smooth, matte cover lies next to a glossy yellow vase. The vase, with a slightly curved silhouette, stands on a dark wood table with a noticeable grain pattern. The book appears slightly worn at the edges, suggesting frequent use, while the vase holds a fresh array of multicolored wildflowers.",
+        ),
+        ("racoon_apple", "a racoon holding a shiny red apple over its head"),
+        (
+            "racoon_apple_Qwen-72B-Chat-refined",
+            "a mischievous raccoon standing on its hind legs, holding a bright red apple aloft in its furry paws. the apple shines brightly against the backdrop of a dense forest, with leaves rustling in the gentle breeze. a few scattered rocks can be seen on the ground beneath the raccoon's feet, while a gnarled tree trunk stands nearby.",
+        ),
+    ]
+
+    # hard example prompt.
+    prompt_name_examples2 = [
+        (
+            "falcon_chinese",
+            "a chinese man wearing a white shirt and a checkered headscarf, holds a large falcon near his shoulder. the falcon has dark feathers with a distinctive beak. the background consists of a clear sky and a fence, suggesting an outdoor setting, possibly a desert or arid region",
+        ),
+        (
+            "wombat",
+            "A close-up photo of a wombat wearing a red backpack and raising both arms in the air. Mount Rushmore is in the background",
+        ),
+        (
+            "bakkot_AstralCodexTen_2",
+            "An oil painting of a man in a factory looking at a cat wearing a top hat",
+        ),
+    ]
+
+    for name, prompt in prompt_name_examples1 + prompt_name_examples2:
+        print("#" * 80)
+        print(f'{name}: "{prompt}"')
+        _batch_size = 1
+        size = 512
+        seed = 1001
+        prompt = [prompt] * _batch_size
+
+        load_ella_for_pipe(pipe, ella)
+        image_flexible = generate_image_with_flexible_max_length(
+            pipe,
+            t5_encoder,
+            prompt,
+            guidance_scale=12,
+            num_inference_steps=50,
+            height=size,
+            width=size,
+            generator=[
+                torch.Generator(device="cuda").manual_seed(seed + i)
+                for i in range(_batch_size)
+            ],
+        )
+        image_fixed = generate_image_with_fixed_max_length(
+            pipe,
+            t5_encoder,
+            prompt,
+            guidance_scale=12,
+            num_inference_steps=50,
+            height=size,
+            width=size,
+            generator=[
+                torch.Generator(device="cuda").manual_seed(seed + i)
+                for i in range(_batch_size)
+            ],
+        )
+        offload_ella_for_pipe(pipe)
+
+        image_ori = pipe(
+            prompt,
+            output_type="pt",
+            guidance_scale=12,
+            num_inference_steps=50,
+            height=size,
+            width=size,
+            generator=[
+                torch.Generator(device="cuda").manual_seed(seed + i)
+                for i in range(_batch_size)
+            ],
+        ).images
+
+        print(f'save image at {save_folder / f"{name}.png"}')
+        print(
+            "original SD1.5\t|\tELLA-SD1.5(fixed token length)\t|\tELLA-SD1.5(flexible token length)"
+        )
+        save_image(
+            torch.cat([image_ori, image_fixed, image_flexible], dim=0),
+            save_folder / f"{name}.png",
+            nrow=3,
+        )
+
+
+if __name__ == "__main__":
+    fire.Fire(dict(test=main, demo=build_demo))

nodes.py

@@ -0,0 +1,261 @@
+import os
+from contextlib import nullcontext
+import torch
+try:
+    from diffusers import (
+        DPMSolverMultistepScheduler, 
+        EulerDiscreteScheduler, 
+        EulerAncestralDiscreteScheduler, 
+        AutoencoderKL, 
+        LCMScheduler, 
+        DDPMScheduler, 
+        DEISMultistepScheduler, 
+        PNDMScheduler,
+        UniPCMultistepScheduler
+    )
+    from diffusers.loaders.single_file_utils import (
+        convert_ldm_vae_checkpoint, 
+        convert_ldm_unet_checkpoint, 
+        create_vae_diffusers_config, 
+        create_unet_diffusers_config,
+        create_text_encoder_from_ldm_clip_checkpoint
+    )            
+except:
+    raise ImportError("Diffusers version too old. Please update to 0.26.0 minimum.")
+
+from .brushnet.pipeline_brushnet import StableDiffusionBrushNetPipeline
+from .brushnet.brushnet import BrushNetModel
+from .brushnet.unet_2d_condition import UNet2DConditionModel
+
+from omegaconf import OmegaConf
+from transformers import CLIPTokenizer
+import comfy.model_management as mm
+import comfy.utils
+import folder_paths
+
+script_directory = os.path.dirname(os.path.abspath(__file__))
+
+class brushnet_model_loader:
+    @classmethod
+    def INPUT_TYPES(s):
+        return {"required": {
+            "model": ("MODEL",),
+            "clip": ("CLIP",),
+            "vae": ("VAE",),
+            "brushnet_model": (
+                [
+                    "brushnet_segmentation_mask",
+                    "brushnet_random_mask",
+                ], {
+                    "default": "brushnet_segmentation_mask"
+                }),
+            },
+        }
+
+    RETURN_TYPES = ("BRUSHNET",)
+    RETURN_NAMES = ("brushnet",)
+    FUNCTION = "loadmodel"
+    CATEGORY = "BrushNetWrapper"
+
+    def loadmodel(self, model, clip, vae, brushnet_model):
+        mm.soft_empty_cache()
+        dtype = mm.unet_dtype()
+        vae_dtype = mm.vae_dtype()
+        device = mm.get_torch_device()
+
+        custom_config = {
+            "model": model,
+            "vae": vae,
+            "clip": clip,
+            "brushnet_model": brushnet_model
+        }
+        if not hasattr(self, "model") or self.model == None or custom_config != self.current_config:
+            pbar = comfy.utils.ProgressBar(5)
+            self.current_config = custom_config
+
+            original_config = OmegaConf.load(os.path.join(script_directory, f"configs/v1-inference.yaml"))
+            brushnet_config = OmegaConf.load(os.path.join(script_directory, f"configs/brushnet_config.json"))
+
+            brushnet_model_folder = os.path.join(folder_paths.models_dir,"brushnet")
+            checkpoint_path = os.path.join(brushnet_model_folder, f"{brushnet_model}_fp16.safetensors")
+            print(f"Loading BrushNet from {checkpoint_path}")
+            
+            if not os.path.exists(checkpoint_path):
+                print(f"Selected model: {checkpoint_path} not found, downloading...")
+                from huggingface_hub import snapshot_download
+                snapshot_download(repo_id="Kijai/BrushNet-fp16", allow_patterns=[f"*{brushnet_model}*"], local_dir=brushnet_model_folder, local_dir_use_symlinks=False)
+
+            brushnet = BrushNetModel(**brushnet_config)
+            brushnet_sd = comfy.utils.load_torch_file(checkpoint_path)
+            brushnet.load_state_dict(brushnet_sd)
+            brushnet.to(dtype)
+
+            clip_sd = None
+            load_models = [model]
+            load_models.append(clip.load_model())
+            clip_sd = clip.get_sd()
+            
+            comfy.model_management.load_models_gpu(load_models)
+            sd = model.model.state_dict_for_saving(clip_sd, vae.get_sd(), None)
+
+            # 1. vae
+            converted_vae_config = create_vae_diffusers_config(original_config, image_size=512)
+            converted_vae = convert_ldm_vae_checkpoint(sd, converted_vae_config)
+            vae = AutoencoderKL(**converted_vae_config)
+            vae.load_state_dict(converted_vae, strict=False)
+
+            pbar.update(1)
+            # 2. unet
+            converted_unet_config = create_unet_diffusers_config(original_config, image_size=512)
+            converted_unet = convert_ldm_unet_checkpoint(sd, converted_unet_config)
+            
+            unet = UNet2DConditionModel(**converted_unet_config)
+            unet.load_state_dict(converted_unet, strict=False)
+            unet = unet.to(device)
+
+            pbar.update(1)
+            # 3. text_model
+            print("loading text model")
+            text_encoder = create_text_encoder_from_ldm_clip_checkpoint("openai/clip-vit-large-patch14",sd)
+            scheduler_config = {
+                "num_train_timesteps": 1000,
+                "beta_start":    0.00085,
+                "beta_end":      0.012,
+                "beta_schedule": "scaled_linear",
+                "steps_offset": 1
+            }
+            # 4. tokenizer
+            tokenizer_path = os.path.join(script_directory, "configs/tokenizer")
+            tokenizer = CLIPTokenizer.from_pretrained(tokenizer_path)
+
+            scheduler=DPMSolverMultistepScheduler(**scheduler_config)
+            pbar.update(1)
+            del sd
+
+            pbar.update(1)
+
+            print("creating pipeline")
+          
+            self.pipe = StableDiffusionBrushNetPipeline(
+                unet=unet, 
+                vae=vae, 
+                text_encoder=text_encoder, 
+                tokenizer=tokenizer, 
+                scheduler=scheduler,
+                brushnet=brushnet,
+                requires_safety_checker=False, 
+                safety_checker=None,
+                feature_extractor=None
+            )   
+            print("pipeline created")
+            pbar.update(1)
+            
+            brushnet = {
+                "pipe": self.pipe,
+            }
+   
+        return (brushnet,)
+    
+class brushnet_sampler:
+    @classmethod
+    def INPUT_TYPES(s):
+        return {"required": {
+            "brushnet": ("BRUSHNET",),
+            "image": ("IMAGE",),
+            "mask": ("MASK",),
+            "steps": ("INT", {"default": 25, "min": 1, "max": 200, "step": 1}),
+            "guidance_scale": ("FLOAT", {"default": 1.0, "min": 0.0, "max": 20.0, "step": 0.01}),
+            "seed": ("INT", {"default": 0, "min": 0, "max": 0xffffffffffffffff}),
+            "scheduler": (
+                [
+                    "DPMSolverMultistepScheduler",
+                    "DPMSolverMultistepScheduler_SDE_karras",
+                    "DDPMScheduler",
+                    "LCMScheduler",
+                    "PNDMScheduler",
+                    "DEISMultistepScheduler",
+                    "EulerDiscreteScheduler",
+                    "EulerAncestralDiscreteScheduler",
+                    "UniPCMultistepScheduler"
+                ], {
+                    "default": "UniPCMultistepScheduler"
+                }),
+            "prompt": ("STRING", {"multiline": True, "default": "caption",}),
+           
+            },    
+        }
+
+    RETURN_TYPES = ("IMAGE",)
+    RETURN_NAMES = ("images",)
+    FUNCTION = "process"
+    CATEGORY = "BrushNetWrapper"
+
+    def process(self, brushnet, image, mask, prompt, steps, guidance_scale, seed, scheduler):
+        device = mm.get_torch_device()
+        mm.unload_all_models()
+        mm.soft_empty_cache()
+        dtype = mm.unet_dtype()
+        pipe=brushnet["pipe"]
+        pipe.to(device, dtype=dtype)
+
+        scheduler_config = {
+                "num_train_timesteps": 1000,
+                "beta_start":    0.00085,
+                "beta_end":      0.012,
+                "beta_schedule": "scaled_linear",
+                "steps_offset": 1,
+            }
+        if scheduler == "DPMSolverMultistepScheduler":
+            noise_scheduler = DPMSolverMultistepScheduler(**scheduler_config)
+        elif scheduler == "DPMSolverMultistepScheduler_SDE_karras":
+            scheduler_config.update({"algorithm_type": "sde-dpmsolver++"})
+            scheduler_config.update({"use_karras_sigmas": True})
+            noise_scheduler = DPMSolverMultistepScheduler(**scheduler_config)
+        elif scheduler == "DDPMScheduler":
+            noise_scheduler = DDPMScheduler(**scheduler_config)
+        elif scheduler == "LCMScheduler":
+            noise_scheduler = LCMScheduler(**scheduler_config)
+        elif scheduler == "PNDMScheduler":
+            scheduler_config.update({"set_alpha_to_one": False})
+            scheduler_config.update({"trained_betas": None})
+            noise_scheduler = PNDMScheduler(**scheduler_config)
+        elif scheduler == "DEISMultistepScheduler":
+            noise_scheduler = DEISMultistepScheduler(**scheduler_config)
+        elif scheduler == "EulerDiscreteScheduler":
+            noise_scheduler = EulerDiscreteScheduler(**scheduler_config)
+        elif scheduler == "EulerAncestralDiscreteScheduler":
+            noise_scheduler = EulerAncestralDiscreteScheduler(**scheduler_config)
+        elif scheduler == "UniPCMultistepScheduler":
+            noise_scheduler = UniPCMultistepScheduler(**scheduler_config)
+        pipe.scheduler = noise_scheduler
+
+        
+        image = image.permute(0, 3, 1, 2).to(device)
+        mask = mask.unsqueeze(0).to(device)
+        image = image * (1-mask)
+
+        autocast_condition = (dtype != torch.float32) and not mm.is_device_mps(device)
+        with torch.autocast(mm.get_autocast_device(device), dtype=dtype) if autocast_condition else nullcontext():
+            generator = torch.Generator(device).manual_seed(seed)
+            images = pipe(
+                prompt, 
+                image=image, 
+                mask=mask, 
+                num_inference_steps=steps, 
+                generator=generator,
+                brushnet_conditioning_scale=guidance_scale,
+                output_type="pt"
+            ).images
+
+            image_out = images.permute(0, 2, 3, 1).cpu().float()
+            return (image_out,)
+
+
+NODE_CLASS_MAPPINGS = {
+    "brushnet_model_loader": brushnet_model_loader,
+    "brushnet_sampler": brushnet_sampler,
+}
+NODE_DISPLAY_NAME_MAPPINGS = {
+    "brushnet_model_loader": "BrushNet Model Loader",
+    "brushnet_sampler": "BrushNet Sampler",
+}

requirements.txt

@@ -0,0 +1,2 @@
+diffusers>=0.26.0
+omegaconf