1086 lines
51 KiB
Python
1086 lines
51 KiB
Python
import os
|
|
from typing import List
|
|
|
|
|
|
import torch
|
|
from typing import Optional, Union, Any, Dict, Tuple, List, Callable
|
|
from diffusers.utils.torch_utils import is_compiled_module, is_torch_version, randn_tensor
|
|
from diffusers.utils import (
|
|
USE_PEFT_BACKEND,
|
|
deprecate,
|
|
logging,
|
|
replace_example_docstring,
|
|
scale_lora_layers,
|
|
unscale_lora_layers,
|
|
)
|
|
from diffusers.pipelines.controlnet.pipeline_controlnet import retrieve_timesteps
|
|
from diffusers import StableDiffusionPipeline
|
|
from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion import StableDiffusionPipelineOutput
|
|
from diffusers.pipelines.controlnet.pipeline_controlnet import StableDiffusionControlNetPipeline
|
|
from diffusers.models.controlnet import ControlNetModel
|
|
from diffusers.image_processor import PipelineImageInput
|
|
from diffusers.pipelines.controlnet import MultiControlNetModel
|
|
from PIL import Image
|
|
from safetensors import safe_open
|
|
from transformers import CLIPImageProcessor, CLIPVisionModelWithProjection
|
|
from torchvision import transforms
|
|
from .style_encoder import Style_Aware_Encoder
|
|
from .tools import pre_processing
|
|
|
|
from .utils import is_torch2_available
|
|
|
|
if is_torch2_available():
|
|
from .attention_processor import (
|
|
AttnProcessor2_0 as AttnProcessor,
|
|
)
|
|
from .attention_processor import (
|
|
CNAttnProcessor2_0 as CNAttnProcessor,
|
|
)
|
|
from .attention_processor import (
|
|
IPAttnProcessor2_0 as IPAttnProcessor,
|
|
)
|
|
else:
|
|
from .attention_processor import AttnProcessor, CNAttnProcessor, IPAttnProcessor
|
|
from .resampler import Resampler
|
|
|
|
|
|
class ImageProjModel(torch.nn.Module):
|
|
"""Projection Model"""
|
|
|
|
def __init__(self, cross_attention_dim=1024, clip_embeddings_dim=1024, clip_extra_context_tokens=4):
|
|
super().__init__()
|
|
|
|
self.cross_attention_dim = cross_attention_dim
|
|
self.clip_extra_context_tokens = clip_extra_context_tokens
|
|
self.proj = torch.nn.Linear(clip_embeddings_dim, self.clip_extra_context_tokens * cross_attention_dim)
|
|
self.norm = torch.nn.LayerNorm(cross_attention_dim)
|
|
|
|
def forward(self, image_embeds):
|
|
embeds = image_embeds
|
|
clip_extra_context_tokens = self.proj(embeds).reshape(
|
|
-1, self.clip_extra_context_tokens, self.cross_attention_dim
|
|
)
|
|
clip_extra_context_tokens = self.norm(clip_extra_context_tokens)
|
|
return clip_extra_context_tokens
|
|
|
|
|
|
class MLPProjModel(torch.nn.Module):
|
|
"""SD model with image prompt"""
|
|
def __init__(self, cross_attention_dim=1024, clip_embeddings_dim=1024):
|
|
super().__init__()
|
|
|
|
self.proj = torch.nn.Sequential(
|
|
torch.nn.Linear(clip_embeddings_dim, clip_embeddings_dim),
|
|
torch.nn.GELU(),
|
|
torch.nn.Linear(clip_embeddings_dim, cross_attention_dim),
|
|
torch.nn.LayerNorm(cross_attention_dim)
|
|
)
|
|
|
|
def forward(self, image_embeds):
|
|
clip_extra_context_tokens = self.proj(image_embeds)
|
|
return clip_extra_context_tokens
|
|
|
|
|
|
class IPAdapter:
|
|
def __init__(self, sd_pipe, image_encoder_path, ip_ckpt, device, num_tokens=4):
|
|
self.device = device
|
|
self.image_encoder_path = image_encoder_path
|
|
self.ip_ckpt = ip_ckpt
|
|
self.num_tokens = num_tokens
|
|
|
|
self.pipe = sd_pipe.to(self.device)
|
|
self.set_ip_adapter()
|
|
|
|
# load image encoder
|
|
self.image_encoder = CLIPVisionModelWithProjection.from_pretrained(self.image_encoder_path).to(
|
|
self.device, dtype=torch.float16
|
|
)
|
|
self.clip_image_processor = CLIPImageProcessor()
|
|
# image proj model
|
|
self.image_proj_model = self.init_proj()
|
|
|
|
self.load_ip_adapter()
|
|
|
|
def init_proj(self):
|
|
image_proj_model = ImageProjModel(
|
|
cross_attention_dim=self.pipe.unet.config.cross_attention_dim,
|
|
clip_embeddings_dim=self.image_encoder.config.projection_dim,
|
|
clip_extra_context_tokens=self.num_tokens,
|
|
).to(self.device, dtype=torch.float16)
|
|
return image_proj_model
|
|
|
|
def set_ip_adapter(self):
|
|
unet = self.pipe.unet
|
|
attn_procs = {}
|
|
for name in unet.attn_processors.keys():
|
|
cross_attention_dim = None if name.endswith("attn1.processor") else unet.config.cross_attention_dim
|
|
if name.startswith("mid_block"):
|
|
hidden_size = unet.config.block_out_channels[-1]
|
|
elif name.startswith("up_blocks"):
|
|
block_id = int(name[len("up_blocks.")])
|
|
hidden_size = list(reversed(unet.config.block_out_channels))[block_id]
|
|
elif name.startswith("down_blocks"):
|
|
block_id = int(name[len("down_blocks.")])
|
|
hidden_size = unet.config.block_out_channels[block_id]
|
|
if cross_attention_dim is None:
|
|
attn_procs[name] = AttnProcessor()
|
|
else:
|
|
attn_procs[name] = IPAttnProcessor(
|
|
hidden_size=hidden_size,
|
|
cross_attention_dim=cross_attention_dim,
|
|
scale=1.0,
|
|
num_tokens=self.num_tokens,
|
|
).to(self.device, dtype=torch.float16)
|
|
unet.set_attn_processor(attn_procs)
|
|
if hasattr(self.pipe, "controlnet"):
|
|
if isinstance(self.pipe.controlnet, MultiControlNetModel):
|
|
for controlnet in self.pipe.controlnet.nets:
|
|
controlnet.set_attn_processor(CNAttnProcessor(num_tokens=self.num_tokens))
|
|
else:
|
|
self.pipe.controlnet.set_attn_processor(CNAttnProcessor(num_tokens=self.num_tokens))
|
|
|
|
def load_ip_adapter(self):
|
|
if os.path.splitext(self.ip_ckpt)[-1] == ".safetensors":
|
|
state_dict = {"image_proj": {}, "ip_adapter": {}}
|
|
with safe_open(self.ip_ckpt, framework="pt", device="cpu") as f:
|
|
for key in f.keys():
|
|
if key.startswith("image_proj."):
|
|
state_dict["image_proj"][key.replace("image_proj.", "")] = f.get_tensor(key)
|
|
elif key.startswith("ip_adapter."):
|
|
state_dict["ip_adapter"][key.replace("ip_adapter.", "")] = f.get_tensor(key)
|
|
else:
|
|
state_dict = torch.load(self.ip_ckpt, map_location="cpu")
|
|
self.image_proj_model.load_state_dict(state_dict["image_proj"])
|
|
ip_layers = torch.nn.ModuleList(self.pipe.unet.attn_processors.values())
|
|
ip_layers.load_state_dict(state_dict["ip_adapter"])
|
|
|
|
@torch.inference_mode()
|
|
def get_image_embeds(self, pil_image=None, clip_image_embeds=None):
|
|
if pil_image is not None:
|
|
if isinstance(pil_image, Image.Image):
|
|
pil_image = [pil_image]
|
|
clip_image = self.clip_image_processor(images=pil_image, return_tensors="pt").pixel_values
|
|
clip_image_embeds = self.image_encoder(clip_image.to(self.device, dtype=torch.float16)).image_embeds
|
|
else:
|
|
clip_image_embeds = clip_image_embeds.to(self.device, dtype=torch.float16)
|
|
image_prompt_embeds = self.image_proj_model(clip_image_embeds)
|
|
uncond_image_prompt_embeds = self.image_proj_model(torch.zeros_like(clip_image_embeds))
|
|
return image_prompt_embeds, uncond_image_prompt_embeds
|
|
|
|
def set_scale(self, scale):
|
|
for attn_processor in self.pipe.unet.attn_processors.values():
|
|
if isinstance(attn_processor, IPAttnProcessor):
|
|
attn_processor.scale = scale
|
|
|
|
def generate(
|
|
self,
|
|
pil_image=None,
|
|
clip_image_embeds=None,
|
|
prompt=None,
|
|
negative_prompt=None,
|
|
scale=1.0,
|
|
num_samples=4,
|
|
seed=None,
|
|
guidance_scale=7.5,
|
|
num_inference_steps=30,
|
|
**kwargs,
|
|
):
|
|
self.set_scale(scale)
|
|
|
|
if pil_image is not None:
|
|
num_prompts = 1 if isinstance(pil_image, Image.Image) else len(pil_image)
|
|
else:
|
|
num_prompts = clip_image_embeds.size(0)
|
|
|
|
if prompt is None:
|
|
prompt = "best quality, high quality"
|
|
if negative_prompt is None:
|
|
negative_prompt = "monochrome, lowres, bad anatomy, worst quality, low quality"
|
|
|
|
if not isinstance(prompt, List):
|
|
prompt = [prompt] * num_prompts
|
|
if not isinstance(negative_prompt, List):
|
|
negative_prompt = [negative_prompt] * num_prompts
|
|
|
|
image_prompt_embeds, uncond_image_prompt_embeds = self.get_image_embeds(
|
|
pil_image=pil_image, clip_image_embeds=clip_image_embeds
|
|
)
|
|
bs_embed, seq_len, _ = image_prompt_embeds.shape
|
|
image_prompt_embeds = image_prompt_embeds.repeat(1, num_samples, 1)
|
|
image_prompt_embeds = image_prompt_embeds.view(bs_embed * num_samples, seq_len, -1)
|
|
uncond_image_prompt_embeds = uncond_image_prompt_embeds.repeat(1, num_samples, 1)
|
|
uncond_image_prompt_embeds = uncond_image_prompt_embeds.view(bs_embed * num_samples, seq_len, -1)
|
|
|
|
with torch.inference_mode():
|
|
prompt_embeds_, negative_prompt_embeds_ = self.pipe.encode_prompt(
|
|
prompt,
|
|
device=self.device,
|
|
num_images_per_prompt=num_samples,
|
|
do_classifier_free_guidance=True,
|
|
negative_prompt=negative_prompt,
|
|
)
|
|
prompt_embeds = torch.cat([prompt_embeds_, image_prompt_embeds], dim=1)
|
|
negative_prompt_embeds = torch.cat([negative_prompt_embeds_, uncond_image_prompt_embeds], dim=1)
|
|
|
|
generator = torch.Generator(self.device).manual_seed(seed) if seed is not None else None
|
|
images = self.pipe(
|
|
prompt_embeds=prompt_embeds,
|
|
negative_prompt_embeds=negative_prompt_embeds,
|
|
guidance_scale=guidance_scale,
|
|
num_inference_steps=num_inference_steps,
|
|
generator=generator,
|
|
**kwargs,
|
|
).images
|
|
|
|
return images
|
|
|
|
|
|
class IPAdapterXL(IPAdapter):
|
|
"""SDXL"""
|
|
|
|
def generate(
|
|
self,
|
|
pil_image,
|
|
prompt=None,
|
|
negative_prompt=None,
|
|
scale=1.0,
|
|
num_samples=4,
|
|
seed=None,
|
|
num_inference_steps=30,
|
|
**kwargs,
|
|
):
|
|
self.set_scale(scale)
|
|
|
|
num_prompts = 1 if isinstance(pil_image, Image.Image) else len(pil_image)
|
|
|
|
if prompt is None:
|
|
prompt = "best quality, high quality"
|
|
if negative_prompt is None:
|
|
negative_prompt = "monochrome, lowres, bad anatomy, worst quality, low quality"
|
|
|
|
if not isinstance(prompt, List):
|
|
prompt = [prompt] * num_prompts
|
|
if not isinstance(negative_prompt, List):
|
|
negative_prompt = [negative_prompt] * num_prompts
|
|
|
|
image_prompt_embeds, uncond_image_prompt_embeds = self.get_image_embeds(pil_image)
|
|
bs_embed, seq_len, _ = image_prompt_embeds.shape
|
|
image_prompt_embeds = image_prompt_embeds.repeat(1, num_samples, 1)
|
|
image_prompt_embeds = image_prompt_embeds.view(bs_embed * num_samples, seq_len, -1)
|
|
uncond_image_prompt_embeds = uncond_image_prompt_embeds.repeat(1, num_samples, 1)
|
|
uncond_image_prompt_embeds = uncond_image_prompt_embeds.view(bs_embed * num_samples, seq_len, -1)
|
|
|
|
with torch.inference_mode():
|
|
(
|
|
prompt_embeds,
|
|
negative_prompt_embeds,
|
|
pooled_prompt_embeds,
|
|
negative_pooled_prompt_embeds,
|
|
) = self.pipe.encode_prompt(
|
|
prompt,
|
|
num_images_per_prompt=num_samples,
|
|
do_classifier_free_guidance=True,
|
|
negative_prompt=negative_prompt,
|
|
)
|
|
prompt_embeds = torch.cat([prompt_embeds, image_prompt_embeds], dim=1)
|
|
negative_prompt_embeds = torch.cat([negative_prompt_embeds, uncond_image_prompt_embeds], dim=1)
|
|
|
|
generator = torch.Generator(self.device).manual_seed(seed) if seed is not None else None
|
|
images = self.pipe(
|
|
prompt_embeds=prompt_embeds,
|
|
negative_prompt_embeds=negative_prompt_embeds,
|
|
pooled_prompt_embeds=pooled_prompt_embeds,
|
|
negative_pooled_prompt_embeds=negative_pooled_prompt_embeds,
|
|
num_inference_steps=num_inference_steps,
|
|
generator=generator,
|
|
**kwargs,
|
|
).images
|
|
|
|
return images
|
|
|
|
|
|
class IPAdapterPlus(IPAdapter):
|
|
"""IP-Adapter with fine-grained features"""
|
|
|
|
def init_proj(self):
|
|
image_proj_model = Resampler(
|
|
dim=self.pipe.unet.config.cross_attention_dim,
|
|
depth=4,
|
|
dim_head=64,
|
|
heads=12,
|
|
num_queries=self.num_tokens,
|
|
embedding_dim=self.image_encoder.config.hidden_size,
|
|
output_dim=self.pipe.unet.config.cross_attention_dim,
|
|
ff_mult=4,
|
|
).to(self.device, dtype=torch.float16)
|
|
return image_proj_model
|
|
|
|
@torch.inference_mode()
|
|
def get_image_embeds(self, pil_image=None, clip_image_embeds=None):
|
|
if isinstance(pil_image, Image.Image):
|
|
pil_image = [pil_image]
|
|
clip_image = self.clip_image_processor(images=pil_image, return_tensors="pt").pixel_values
|
|
clip_image = clip_image.to(self.device, dtype=torch.float16)
|
|
clip_image_embeds = self.image_encoder(clip_image, output_hidden_states=True).hidden_states[-2]
|
|
image_prompt_embeds = self.image_proj_model(clip_image_embeds)
|
|
uncond_clip_image_embeds = self.image_encoder(
|
|
torch.zeros_like(clip_image), output_hidden_states=True
|
|
).hidden_states[-2]
|
|
uncond_image_prompt_embeds = self.image_proj_model(uncond_clip_image_embeds)
|
|
return image_prompt_embeds, uncond_image_prompt_embeds
|
|
|
|
|
|
class IPAdapterFull(IPAdapterPlus):
|
|
"""IP-Adapter with full features"""
|
|
|
|
def init_proj(self):
|
|
image_proj_model = MLPProjModel(
|
|
cross_attention_dim=self.pipe.unet.config.cross_attention_dim,
|
|
clip_embeddings_dim=self.image_encoder.config.hidden_size,
|
|
).to(self.device, dtype=torch.float16)
|
|
return image_proj_model
|
|
|
|
|
|
class IPAdapterPlusXL(IPAdapter):
|
|
"""SDXL"""
|
|
|
|
def init_proj(self):
|
|
image_proj_model = Resampler(
|
|
dim=1280,
|
|
depth=4,
|
|
dim_head=64,
|
|
heads=20,
|
|
num_queries=self.num_tokens,
|
|
embedding_dim=self.image_encoder.config.hidden_size,
|
|
output_dim=self.pipe.unet.config.cross_attention_dim,
|
|
ff_mult=4,
|
|
).to(self.device, dtype=torch.float16)
|
|
return image_proj_model
|
|
|
|
@torch.inference_mode()
|
|
def get_image_embeds(self, pil_image):
|
|
if isinstance(pil_image, Image.Image):
|
|
pil_image = [pil_image]
|
|
clip_image = self.clip_image_processor(images=pil_image, return_tensors="pt").pixel_values
|
|
clip_image = clip_image.to(self.device, dtype=torch.float16)
|
|
clip_image_embeds = self.image_encoder(clip_image, output_hidden_states=True).hidden_states[-2]
|
|
image_prompt_embeds = self.image_proj_model(clip_image_embeds)
|
|
uncond_clip_image_embeds = self.image_encoder(
|
|
torch.zeros_like(clip_image), output_hidden_states=True
|
|
).hidden_states[-2]
|
|
uncond_image_prompt_embeds = self.image_proj_model(uncond_clip_image_embeds)
|
|
return image_prompt_embeds, uncond_image_prompt_embeds
|
|
|
|
def generate(
|
|
self,
|
|
pil_image,
|
|
prompt=None,
|
|
negative_prompt=None,
|
|
scale=1.0,
|
|
num_samples=4,
|
|
seed=None,
|
|
num_inference_steps=30,
|
|
**kwargs,
|
|
):
|
|
self.set_scale(scale)
|
|
|
|
num_prompts = 1 if isinstance(pil_image, Image.Image) else len(pil_image)
|
|
|
|
if prompt is None:
|
|
prompt = "best quality, high quality"
|
|
if negative_prompt is None:
|
|
negative_prompt = "monochrome, lowres, bad anatomy, worst quality, low quality"
|
|
|
|
if not isinstance(prompt, List):
|
|
prompt = [prompt] * num_prompts
|
|
if not isinstance(negative_prompt, List):
|
|
negative_prompt = [negative_prompt] * num_prompts
|
|
|
|
image_prompt_embeds, uncond_image_prompt_embeds = self.get_image_embeds(pil_image)
|
|
bs_embed, seq_len, _ = image_prompt_embeds.shape
|
|
image_prompt_embeds = image_prompt_embeds.repeat(1, num_samples, 1)
|
|
image_prompt_embeds = image_prompt_embeds.view(bs_embed * num_samples, seq_len, -1)
|
|
uncond_image_prompt_embeds = uncond_image_prompt_embeds.repeat(1, num_samples, 1)
|
|
uncond_image_prompt_embeds = uncond_image_prompt_embeds.view(bs_embed * num_samples, seq_len, -1)
|
|
|
|
with torch.inference_mode():
|
|
(
|
|
prompt_embeds,
|
|
negative_prompt_embeds,
|
|
pooled_prompt_embeds,
|
|
negative_pooled_prompt_embeds,
|
|
) = self.pipe.encode_prompt(
|
|
prompt,
|
|
num_images_per_prompt=num_samples,
|
|
do_classifier_free_guidance=True,
|
|
negative_prompt=negative_prompt,
|
|
)
|
|
prompt_embeds = torch.cat([prompt_embeds, image_prompt_embeds], dim=1)
|
|
negative_prompt_embeds = torch.cat([negative_prompt_embeds, uncond_image_prompt_embeds], dim=1)
|
|
|
|
generator = torch.Generator(self.device).manual_seed(seed) if seed is not None else None
|
|
images = self.pipe(
|
|
prompt_embeds=prompt_embeds,
|
|
negative_prompt_embeds=negative_prompt_embeds,
|
|
pooled_prompt_embeds=pooled_prompt_embeds,
|
|
negative_pooled_prompt_embeds=negative_pooled_prompt_embeds,
|
|
num_inference_steps=num_inference_steps,
|
|
generator=generator,
|
|
**kwargs,
|
|
).images
|
|
|
|
return images
|
|
|
|
|
|
def StyleProcessor(style_image, device):
|
|
transform = transforms.Compose([
|
|
transforms.ToTensor(),
|
|
transforms.Normalize([0.5], [0.5]),
|
|
])
|
|
# centercrop for style condition
|
|
crop = transforms.Compose(
|
|
[
|
|
transforms.Resize(512, interpolation=transforms.InterpolationMode.BILINEAR),
|
|
transforms.CenterCrop(512),
|
|
]
|
|
)
|
|
style_image = crop(style_image)
|
|
high_style_patch, middle_style_patch, low_style_patch = pre_processing(style_image.convert("RGB"), transform)
|
|
# shuffling
|
|
high_style_patch, middle_style_patch, low_style_patch = (high_style_patch[torch.randperm(high_style_patch.shape[0])],
|
|
middle_style_patch[torch.randperm(middle_style_patch.shape[0])],
|
|
low_style_patch[torch.randperm(low_style_patch.shape[0])])
|
|
return (high_style_patch.to(device, dtype=torch.float32), middle_style_patch.to(device, dtype=torch.float32), low_style_patch.to(device, dtype=torch.float32))
|
|
|
|
|
|
class StyleShot(torch.nn.Module):
|
|
"""StyleShot generation"""
|
|
def __init__(self, device, pipe, ip_ckpt, style_aware_encoder_ckpt, transformer_patch):
|
|
super().__init__()
|
|
self.num_tokens = 6
|
|
self.device = device
|
|
self.pipe = pipe
|
|
|
|
self.set_ip_adapter(device)
|
|
self.ip_ckpt = ip_ckpt
|
|
|
|
self.style_aware_encoder = Style_Aware_Encoder(CLIPVisionModelWithProjection.from_pretrained(transformer_patch)).to(self.device, dtype=torch.float32)
|
|
self.style_aware_encoder.load_state_dict(torch.load(style_aware_encoder_ckpt))
|
|
|
|
self.style_image_proj_modules = self.init_proj()
|
|
|
|
self.load_ip_adapter()
|
|
self.pipe = self.pipe.to(self.device, dtype=torch.float32)
|
|
|
|
def init_proj(self):
|
|
style_image_proj_modules = torch.nn.ModuleList([
|
|
ImageProjModel(
|
|
cross_attention_dim=self.pipe.unet.config.cross_attention_dim,
|
|
clip_embeddings_dim=self.style_aware_encoder.projection_dim,
|
|
clip_extra_context_tokens=2,
|
|
),
|
|
ImageProjModel(
|
|
cross_attention_dim=self.pipe.unet.config.cross_attention_dim,
|
|
clip_embeddings_dim=self.style_aware_encoder.projection_dim,
|
|
clip_extra_context_tokens=2,
|
|
),
|
|
ImageProjModel(
|
|
cross_attention_dim=self.pipe.unet.config.cross_attention_dim,
|
|
clip_embeddings_dim=self.style_aware_encoder.projection_dim,
|
|
clip_extra_context_tokens=2,
|
|
)])
|
|
return style_image_proj_modules.to(self.device, dtype=torch.float32)
|
|
|
|
def load_ip_adapter(self):
|
|
sd = torch.load(self.ip_ckpt, map_location="cpu")
|
|
style_image_proj_sd = {}
|
|
ip_sd = {}
|
|
controlnet_sd = {}
|
|
for k in sd:
|
|
if k.startswith("unet"):
|
|
pass
|
|
elif k.startswith("style_image_proj_modules"):
|
|
style_image_proj_sd[k.replace("style_image_proj_modules.", "")] = sd[k]
|
|
elif k.startswith("adapter_modules"):
|
|
ip_sd[k.replace("adapter_modules.", "")] = sd[k]
|
|
elif k.startswith("controlnet"):
|
|
controlnet_sd[k.replace("controlnet.", "")] = sd[k]
|
|
# Load state dict for image_proj_model and adapter_modules
|
|
self.style_image_proj_modules.load_state_dict(style_image_proj_sd, strict=True)
|
|
ip_layers = torch.nn.ModuleList(self.pipe.unet.attn_processors.values())
|
|
if hasattr(self.pipe, "controlnet") and isinstance(self.pipe, StyleContentStableDiffusionControlNetPipeline):
|
|
self.pipe.controlnet.load_state_dict(controlnet_sd, strict=True)
|
|
ip_layers.load_state_dict(ip_sd, strict=True)
|
|
|
|
def set_ip_adapter(self, device):
|
|
unet = self.pipe.unet
|
|
attn_procs = {}
|
|
for name in unet.attn_processors.keys():
|
|
cross_attention_dim = None if name.endswith("attn1.processor") else unet.config.cross_attention_dim
|
|
if name.startswith("mid_block"):
|
|
hidden_size = unet.config.block_out_channels[-1]
|
|
elif name.startswith("up_blocks"):
|
|
block_id = int(name[len("up_blocks.")])
|
|
hidden_size = list(reversed(unet.config.block_out_channels))[block_id]
|
|
elif name.startswith("down_blocks"):
|
|
block_id = int(name[len("down_blocks.")])
|
|
hidden_size = unet.config.block_out_channels[block_id]
|
|
if cross_attention_dim is None:
|
|
attn_procs[name] = AttnProcessor()
|
|
else:
|
|
attn_procs[name] = IPAttnProcessor(
|
|
hidden_size=hidden_size,
|
|
cross_attention_dim=cross_attention_dim,
|
|
scale=1.0,
|
|
num_tokens=self.num_tokens,
|
|
).to(device, dtype=torch.float16)
|
|
if hasattr(self.pipe, "controlnet") and not isinstance(self.pipe, StyleContentStableDiffusionControlNetPipeline):
|
|
if isinstance(self.pipe.controlnet, MultiControlNetModel):
|
|
for controlnet in self.pipe.controlnet.nets:
|
|
controlnet.set_attn_processor(CNAttnProcessor(num_tokens=self.num_tokens))
|
|
else:
|
|
self.pipe.controlnet.set_attn_processor(CNAttnProcessor(num_tokens=self.num_tokens))
|
|
unet.set_attn_processor(attn_procs)
|
|
|
|
@torch.inference_mode()
|
|
def get_image_embeds(self, style_image=None):
|
|
style_image = StyleProcessor(style_image, self.device)
|
|
style_embeds = self.style_aware_encoder(style_image).to(self.device, dtype=torch.float32)
|
|
style_ip_tokens = []
|
|
uncond_style_ip_tokens = []
|
|
for idx, style_embed in enumerate([style_embeds[:, 0, :], style_embeds[:, 1, :], style_embeds[:, 2, :]]):
|
|
style_ip_tokens.append(self.style_image_proj_modules[idx](style_embed))
|
|
uncond_style_ip_tokens.append(self.style_image_proj_modules[idx](torch.zeros_like(style_embed)))
|
|
style_ip_tokens = torch.cat(style_ip_tokens, dim=1)
|
|
uncond_style_ip_tokens = torch.cat(uncond_style_ip_tokens, dim=1)
|
|
return style_ip_tokens, uncond_style_ip_tokens
|
|
|
|
def set_scale(self, scale):
|
|
for attn_processor in self.pipe.unet.attn_processors.values():
|
|
if isinstance(attn_processor, IPAttnProcessor):
|
|
attn_processor.scale = scale
|
|
|
|
def samples(self, image_prompt_embeds, uncond_image_prompt_embeds, num_samples, device, prompt, negative_prompt,
|
|
seed, guidance_scale, num_inference_steps, content_image, **kwargs, ):
|
|
bs_embed, seq_len, _ = image_prompt_embeds.shape
|
|
image_prompt_embeds = image_prompt_embeds.repeat(1, num_samples, 1)
|
|
image_prompt_embeds = image_prompt_embeds.view(bs_embed * num_samples, seq_len, -1)
|
|
uncond_image_prompt_embeds = uncond_image_prompt_embeds.repeat(1, num_samples, 1)
|
|
uncond_image_prompt_embeds = uncond_image_prompt_embeds.view(bs_embed * num_samples, seq_len, -1)
|
|
with torch.inference_mode():
|
|
prompt_embeds_, negative_prompt_embeds_ = self.pipe.encode_prompt(
|
|
prompt,
|
|
device=device,
|
|
num_images_per_prompt=num_samples,
|
|
do_classifier_free_guidance=True,
|
|
negative_prompt=negative_prompt,
|
|
)
|
|
prompt_embeds = torch.cat([prompt_embeds_, image_prompt_embeds], dim=1)
|
|
negative_prompt_embeds = torch.cat([negative_prompt_embeds_, uncond_image_prompt_embeds], dim=1)
|
|
|
|
generator = torch.Generator(device).manual_seed(seed) if seed is not None else None
|
|
if content_image is None:
|
|
images = self.pipe(
|
|
prompt_embeds=prompt_embeds,
|
|
negative_prompt_embeds=negative_prompt_embeds,
|
|
guidance_scale=guidance_scale,
|
|
num_inference_steps=num_inference_steps,
|
|
generator=generator,
|
|
**kwargs,
|
|
).images
|
|
else:
|
|
images = self.pipe(
|
|
prompt_embeds=prompt_embeds,
|
|
negative_prompt_embeds=negative_prompt_embeds,
|
|
guidance_scale=guidance_scale,
|
|
num_inference_steps=num_inference_steps,
|
|
generator=generator,
|
|
image=content_image,
|
|
style_embeddings=image_prompt_embeds,
|
|
negative_style_embeddings=uncond_image_prompt_embeds,
|
|
**kwargs,
|
|
).images
|
|
return images
|
|
|
|
def generate(
|
|
self,
|
|
style_image=None,
|
|
prompt=None,
|
|
negative_prompt=None,
|
|
scale=1.0,
|
|
num_samples=1,
|
|
seed=42,
|
|
guidance_scale=7.5,
|
|
num_inference_steps=50,
|
|
content_image=None,
|
|
**kwargs,
|
|
):
|
|
self.set_scale(scale)
|
|
|
|
num_prompts = 1
|
|
|
|
if prompt is None:
|
|
prompt = "best quality, high quality"
|
|
if negative_prompt is None:
|
|
negative_prompt = "monochrome, lowres, bad anatomy, worst quality, low quality"
|
|
|
|
if not isinstance(prompt, List):
|
|
prompt = [prompt] * num_prompts
|
|
if not isinstance(negative_prompt, List):
|
|
negative_prompt = [negative_prompt] * num_prompts
|
|
|
|
style_ip_tokens, uncond_style_ip_tokens = self.get_image_embeds(style_image)
|
|
generate_images = []
|
|
for p in prompt:
|
|
images = self.samples(style_ip_tokens, uncond_style_ip_tokens, num_samples, self.device, p * num_prompts, negative_prompt, seed, guidance_scale, num_inference_steps, content_image, **kwargs, )
|
|
generate_images.append(images)
|
|
return generate_images
|
|
|
|
|
|
class StyleContentStableDiffusionControlNetPipeline(StableDiffusionControlNetPipeline):
|
|
@torch.no_grad()
|
|
def __call__(
|
|
self,
|
|
prompt: Union[str, List[str]] = None,
|
|
image: 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,
|
|
controlnet_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"],
|
|
style_embeddings: Optional[torch.FloatTensor] = None,
|
|
negative_style_embeddings: Optional[torch.FloatTensor] = 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 ControlNet input condition to provide guidance to the `unet` for generation. If the type is
|
|
specified as `torch.FloatTensor`, it is passed to ControlNet 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 ControlNets 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 ControlNet. When `prompt` is a list, and if a list of images is passed for a single ControlNet,
|
|
each will be paired with each prompt in the `prompt` list. This also applies to multiple ControlNets,
|
|
where a list of image lists can be passed to batch for each prompt and each ControlNet.
|
|
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).
|
|
controlnet_conditioning_scale (`float` or `List[float]`, *optional*, defaults to 1.0):
|
|
The outputs of the ControlNet are multiplied by `controlnet_conditioning_scale` before they are added
|
|
to the residual in the original `unet`. If multiple ControlNets are specified in `init`, you can set
|
|
the corresponding scale as a list.
|
|
guess_mode (`bool`, *optional*, defaults to `False`):
|
|
The ControlNet 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 ControlNet starts applying.
|
|
control_guidance_end (`float` or `List[float]`, *optional*, defaults to 1.0):
|
|
The percentage of total steps at which the ControlNet 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`",
|
|
)
|
|
|
|
controlnet = self.controlnet._orig_mod if is_compiled_module(self.controlnet) else self.controlnet
|
|
|
|
# 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(controlnet.nets) if isinstance(controlnet, MultiControlNetModel) else 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,
|
|
image,
|
|
callback_steps,
|
|
negative_prompt,
|
|
prompt_embeds,
|
|
negative_prompt_embeds,
|
|
ip_adapter_image,
|
|
ip_adapter_image_embeds,
|
|
controlnet_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
|
|
|
|
if isinstance(controlnet, MultiControlNetModel) and isinstance(controlnet_conditioning_scale, float):
|
|
controlnet_conditioning_scale = [controlnet_conditioning_scale] * len(controlnet.nets)
|
|
|
|
global_pool_conditions = (
|
|
controlnet.config.global_pool_conditions
|
|
if isinstance(controlnet, ControlNetModel)
|
|
else controlnet.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(controlnet, ControlNetModel):
|
|
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=controlnet.dtype,
|
|
do_classifier_free_guidance=self.do_classifier_free_guidance,
|
|
guess_mode=guess_mode,
|
|
)
|
|
height, width = image.shape[-2:]
|
|
|
|
elif isinstance(controlnet, MultiControlNetModel):
|
|
images = []
|
|
|
|
# Nested lists as ControlNet condition
|
|
if isinstance(image[0], list):
|
|
# Transpose the nested image list
|
|
image = [list(t) for t in zip(*image)]
|
|
|
|
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=controlnet.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
|
|
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 = self.prepare_latents(
|
|
batch_size * num_images_per_prompt,
|
|
num_channels_latents,
|
|
height,
|
|
width,
|
|
prompt_embeds.dtype,
|
|
device,
|
|
generator,
|
|
latents,
|
|
)
|
|
|
|
# 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 controlnets to keep
|
|
controlnet_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)
|
|
]
|
|
controlnet_keep.append(keeps[0] if isinstance(controlnet, ControlNetModel) 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_controlnet_compiled = is_compiled_module(self.controlnet)
|
|
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_controlnet_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)
|
|
|
|
# controlnet(s) inference
|
|
if guess_mode and self.do_classifier_free_guidance:
|
|
# Infer ControlNet only for the conditional batch.
|
|
control_model_input = latents
|
|
control_model_input = self.scheduler.scale_model_input(control_model_input, t)
|
|
controlnet_prompt_embeds = prompt_embeds.chunk(2)[1]
|
|
else:
|
|
control_model_input = latent_model_input
|
|
controlnet_prompt_embeds = prompt_embeds
|
|
|
|
if isinstance(controlnet_keep[i], list):
|
|
cond_scale = [c * s for c, s in zip(controlnet_conditioning_scale, controlnet_keep[i])]
|
|
else:
|
|
controlnet_cond_scale = controlnet_conditioning_scale
|
|
if isinstance(controlnet_cond_scale, list):
|
|
controlnet_cond_scale = controlnet_cond_scale[0]
|
|
cond_scale = controlnet_cond_scale * controlnet_keep[i]
|
|
|
|
if self.do_classifier_free_guidance:
|
|
style_embeddings_input = torch.cat([negative_style_embeddings, style_embeddings])
|
|
|
|
down_block_res_samples, mid_block_res_sample = self.controlnet(
|
|
control_model_input,
|
|
t,
|
|
encoder_hidden_states=style_embeddings_input,
|
|
controlnet_cond=image,
|
|
conditioning_scale=cond_scale,
|
|
guess_mode=guess_mode,
|
|
return_dict=False,
|
|
)
|
|
|
|
if guess_mode and self.do_classifier_free_guidance:
|
|
# Infered ControlNet only for the conditional batch.
|
|
# To apply the output of ControlNet 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])
|
|
|
|
# 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_additional_residuals=down_block_res_samples,
|
|
mid_block_additional_residual=mid_block_res_sample,
|
|
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 controlnet
|
|
# manually for max memory savings
|
|
if hasattr(self, "final_offload_hook") and self.final_offload_hook is not None:
|
|
self.unet.to("cpu")
|
|
self.controlnet.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) |