import inspect from PIL import Image import os import torch from torch import autocast from torchvision import transforms as T from types import MethodType from typing import List, Optional, Tuple, Union from diffusers import StableDiffusionPipeline from diffusers.models.unet_2d_condition import UNet2DConditionOutput from diffusers.pipelines.stable_diffusion import StableDiffusionPipelineOutput COLAB = False if COLAB: pipe = StableDiffusionPipeline.from_pretrained("CompVis/stable-diffusion-v1-4", revision="fp16", torch_dtype=torch.float16, use_auth_token=os.environ.get('HF_TOKEN_SD')) else: pipe = StableDiffusionPipeline.from_pretrained("CompVis/stable-diffusion-v1-4", use_auth_token=os.environ.get('HF_TOKEN_SD')) #pipe = pipe.to('cuda') # Overriding the U-Net forward pass def forward( self, sample: torch.FloatTensor, timestep: Union[torch.Tensor, float, int], encoder_hidden_states: torch.Tensor, return_dict: bool = True, ) -> Union[UNet2DConditionOutput, Tuple]: """r Args: sample (`torch.FloatTensor`): (batch, channel, height, width) noisy inputs tensor timestep (`torch.FloatTensor` or `float` or `int`): (batch) timesteps encoder_hidden_states (`torch.FloatTensor`): (batch, channel, height, width) encoder hidden states return_dict (`bool`, *optional*, defaults to `True`): Whether or not to return a [`models.unet_2d_condition.UNet2DConditionOutput`] instead of a plain tuple. Returns: [`~models.unet_2d_condition.UNet2DConditionOutput`] or `tuple`: [`~models.unet_2d_condition.UNet2DConditionOutput`] if `return_dict` is True, otherwise a `tuple`. When returning a tuple, the first element is the sample tensor. """ # 0. center input if necessary if self.config.center_input_sample: sample = 2 * sample - 1.0 # 1. time timesteps = timestep if not torch.is_tensor(timesteps): timesteps = torch.tensor([timesteps], dtype=torch.long, device=sample.device) elif torch.is_tensor(timesteps) and len(timesteps.shape) == 0: timesteps = timesteps.to(dtype=torch.float32) timesteps = timesteps[None].to(device=sample.device) # broadcast to batch dimension in a way that's compatible with ONNX/Core ML timesteps = timesteps.expand(sample.shape[0]) t_emb = self.time_proj(timesteps) #emb = self.time_embedding(t_emb) emb = self.time_embedding(t_emb.to(sample.dtype)) # 2. pre-process sample = self.conv_in(sample) # 3. down down_block_res_samples = (sample,) for downsample_block in self.down_blocks: if hasattr(downsample_block, "attentions") and downsample_block.attentions is not None: sample, res_samples = downsample_block( hidden_states=sample, temb=emb, encoder_hidden_states=encoder_hidden_states ) else: sample, res_samples = downsample_block(hidden_states=sample, temb=emb) down_block_res_samples += res_samples # 4. mid sample = self.mid_block(sample, emb, encoder_hidden_states=encoder_hidden_states) # 5. up for upsample_block in self.up_blocks: res_samples = down_block_res_samples[-len(upsample_block.resnets) :] down_block_res_samples = down_block_res_samples[: -len(upsample_block.resnets)] if hasattr(upsample_block, "attentions") and upsample_block.attentions is not None: sample = upsample_block( hidden_states=sample, temb=emb, res_hidden_states_tuple=res_samples, encoder_hidden_states=encoder_hidden_states, ) else: sample = upsample_block(hidden_states=sample, temb=emb, res_hidden_states_tuple=res_samples) # 6. post-process # make sure hidden states is in float32 # when running in half-precision #sample = self.conv_norm_out(sample.float()).type(sample.dtype) sample = self.conv_norm_out(sample) sample = self.conv_act(sample) sample = self.conv_out(sample) if not return_dict: return (sample,) return UNet2DConditionOutput(sample=sample) def safety_forward(self, clip_input, images): return images, False # Overriding the Stable Diffusion call method @torch.no_grad() def call( self, prompt: Union[str, List[str]], height: Optional[int] = 512, width: Optional[int] = 512, num_inference_steps: Optional[int] = 50, guidance_scale: Optional[float] = 7.5, eta: Optional[float] = 0.0, generator: Optional[torch.Generator] = None, latents: Optional[torch.FloatTensor] = None, output_type: Optional[str] = "pil", return_dict: bool = True, percent_noise: float = 0.7, **kwargs, ): if isinstance(prompt, str): batch_size = 1 elif isinstance(prompt, list): batch_size = len(prompt) else: raise ValueError(f"`prompt` has to be of type `str` or `list` but is {type(prompt)}") if height % 8 != 0 or width % 8 != 0: raise ValueError(f"`height` and `width` have to be divisible by 8 but are {height} and {width}.") # get prompt text embeddings text_input = self.tokenizer( prompt, padding="max_length", max_length=self.tokenizer.model_max_length, truncation=True, return_tensors="pt", ) text_embeddings = self.text_encoder(text_input.input_ids.to(self.device))[0] # 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. do_classifier_free_guidance = guidance_scale > 1.0 # get unconditional embeddings for classifier free guidance if do_classifier_free_guidance: max_length = text_input.input_ids.shape[-1] uncond_input = self.tokenizer( [""] * batch_size, padding="max_length", max_length=max_length, return_tensors="pt" ) uncond_embeddings = self.text_encoder(uncond_input.input_ids.to(self.device))[0] # 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 text_embeddings = torch.cat([uncond_embeddings, text_embeddings]) # get the initial random noise unless the user supplied it # Unlike in other pipelines, latents need to be generated in the target device # for 1-to-1 results reproducibility with the CompVis implementation. # However this currently doesn't work in `mps`. latents_device = "cpu" if self.device.type == "mps" else self.device latents_shape = (batch_size, self.unet.in_channels, height // 8, width // 8) if latents is None: latents = torch.randn( latents_shape, generator=generator, device=latents_device, ) else: if latents.shape != latents_shape: raise ValueError(f"Unexpected latents shape, got {latents.shape}, expected {latents_shape}") latents = latents.to(self.device) # set timesteps self.scheduler.set_timesteps(num_inference_steps) # if we use LMSDiscreteScheduler, let's make sure latents are multiplied by sigmas #if isinstance(self.scheduler, LMSDiscreteScheduler): # latents = latents * self.scheduler.sigmas[0] # 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 for i, t in enumerate(self.progress_bar(self.scheduler.timesteps)): if t - 1 > 1000 * percent_noise: continue #print(t) # expand the latents if we are doing classifier free guidance latent_model_input = torch.cat([latents] * 2) if do_classifier_free_guidance else latents #if isinstance(self.scheduler, LMSDiscreteScheduler): # sigma = self.scheduler.sigmas[i] # # the model input needs to be scaled to match the continuous ODE formulation in K-LMS # latent_model_input = latent_model_input / ((sigma**2 + 1) ** 0.5) # predict the noise residual noise_pred = self.unet(latent_model_input, t, encoder_hidden_states=text_embeddings).sample # perform guidance if 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 #if isinstance(self.scheduler, LMSDiscreteScheduler): # latents = self.scheduler.step(noise_pred, i, latents, **extra_step_kwargs).prev_sample #else: latents = self.scheduler.step(noise_pred, t, latents, **extra_step_kwargs).prev_sample # scale and decode the image latents with vae latents = 1 / 0.18215 * latents image = self.vae.decode(latents).sample image = (image / 2 + 0.5).clamp(0, 1) image = image.cpu().permute(0, 2, 3, 1).numpy() # run safety checker safety_cheker_input = self.feature_extractor(self.numpy_to_pil(image), return_tensors="pt").to(self.device) image, has_nsfw_concept = self.safety_checker(images=image, clip_input=safety_cheker_input.pixel_values) if output_type == "pil": image = self.numpy_to_pil(image) if not return_dict: return (image, has_nsfw_concept) return StableDiffusionPipelineOutput(images=image, nsfw_content_detected=has_nsfw_concept) if COLAB: pipe.unet.forward = MethodType(forward, pipe.unet) pipe.safety_checker.forward = MethodType(safety_forward, pipe.safety_checker) type(pipe).__call__ = call