Finalize stable diffusion refactor (#1269)

* finish

* cleaner

* more fixes

* refactor

* make fix copies

* refactor cycle diffusion

* finish

* finish2

* Apply suggestions from code review

src/diffusers/pipelines/stable_diffusion/pipeline_cycle_diffusion.py

@@ -19,6 +19,7 @@ import numpy as np
 import torch
 
 import PIL
+from diffusers.utils import is_accelerate_available
 from transformers import CLIPFeatureExtractor, CLIPTextModel, CLIPTokenizer
 
 from ...configuration_utils import FrozenDict
@@ -178,6 +179,7 @@ class CycleDiffusionPipeline(DiffusionPipeline):
             feature_extractor=feature_extractor,
         )
 
+    # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.enable_attention_slicing
     def enable_attention_slicing(self, slice_size: Optional[Union[str, int]] = "auto"):
         r"""
         Enable sliced attention computation.
@@ -197,14 +199,33 @@ class CycleDiffusionPipeline(DiffusionPipeline):
             slice_size = self.unet.config.attention_head_dim // 2
         self.unet.set_attention_slice(slice_size)
 
+    # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.disable_attention_slicing
     def disable_attention_slicing(self):
         r"""
         Disable sliced attention computation. If `enable_attention_slicing` was previously invoked, this method will go
         back to computing attention in one step.
         """
-        # set slice_size = `None` to disable `set_attention_slice`
+        # set slice_size = `None` to disable `attention slicing`
         self.enable_attention_slicing(None)
 
+    # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.enable_sequential_cpu_offload
+    def enable_sequential_cpu_offload(self):
+        r"""
+        Offloads all models to CPU using accelerate, significantly reducing memory usage. When called, unet,
+        text_encoder, vae and safety checker have their state dicts saved to CPU and then are moved to a
+        `torch.device('meta') and loaded to GPU only when their specific submodule has its `forward` method called.
+        """
+        if is_accelerate_available():
+            from accelerate import cpu_offload
+        else:
+            raise ImportError("Please install accelerate via `pip install accelerate`")
+
+        device = torch.device("cuda")
+
+        for cpu_offloaded_model in [self.unet, self.text_encoder, self.vae, self.safety_checker]:
+            if cpu_offloaded_model is not None:
+                cpu_offload(cpu_offloaded_model, device)
+
     @property
     # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline._execution_device
     def _execution_device(self):
@@ -224,6 +245,26 @@ class CycleDiffusionPipeline(DiffusionPipeline):
                 return torch.device(module._hf_hook.execution_device)
         return self.device
 
+    # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.enable_xformers_memory_efficient_attention
+    def enable_xformers_memory_efficient_attention(self):
+        r"""
+        Enable memory efficient attention as implemented in xformers.
+
+        When this option is enabled, you should observe lower GPU memory usage and a potential speed up at inference
+        time. Speed up at training time is not guaranteed.
+
+        Warning: When Memory Efficient Attention and Sliced attention are both enabled, the Memory Efficient Attention
+        is used.
+        """
+        self.unet.set_use_memory_efficient_attention_xformers(True)
+
+    # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.disable_xformers_memory_efficient_attention
+    def disable_xformers_memory_efficient_attention(self):
+        r"""
+        Disable memory efficient attention as implemented in xformers.
+        """
+        self.unet.set_use_memory_efficient_attention_xformers(False)
+
     # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline._encode_prompt
     def _encode_prompt(self, prompt, device, num_images_per_prompt, do_classifier_free_guidance, negative_prompt):
         r"""
@@ -310,6 +351,106 @@ class CycleDiffusionPipeline(DiffusionPipeline):
 
         return text_embeddings
 
+    # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion_img2img.StableDiffusionImg2ImgPipeline.check_inputs
+    def check_inputs(self, prompt, strength, callback_steps):
+        if not isinstance(prompt, str) and not isinstance(prompt, list):
+            raise ValueError(f"`prompt` has to be of type `str` or `list` but is {type(prompt)}")
+
+        if strength < 0 or strength > 1:
+            raise ValueError(f"The value of strength should in [1.0, 1.0] but is {strength}")
+
+        if (callback_steps is None) or (
+            callback_steps is not None and (not isinstance(callback_steps, int) or callback_steps <= 0)
+        ):
+            raise ValueError(
+                f"`callback_steps` has to be a positive integer but is {callback_steps} of type"
+                f" {type(callback_steps)}."
+            )
+
+    # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.prepare_extra_step_kwargs
+    def prepare_extra_step_kwargs(self, generator, eta):
+        # prepare extra kwargs for the scheduler step, since not all schedulers have the same signature
+        # eta (η) is only used with the DDIMScheduler, it will be ignored for other schedulers.
+        # eta corresponds to η in DDIM paper: https://arxiv.org/abs/2010.02502
+        # and should be between [0, 1]
+
+        accepts_eta = "eta" in set(inspect.signature(self.scheduler.step).parameters.keys())
+        extra_step_kwargs = {}
+        if accepts_eta:
+            extra_step_kwargs["eta"] = eta
+
+        # check if the scheduler accepts generator
+        accepts_generator = "generator" in set(inspect.signature(self.scheduler.step).parameters.keys())
+        if accepts_generator:
+            extra_step_kwargs["generator"] = generator
+        return extra_step_kwargs
+
+    # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.run_safety_checker
+    def run_safety_checker(self, image, device, dtype):
+        if self.safety_checker is not None:
+            safety_checker_input = self.feature_extractor(self.numpy_to_pil(image), return_tensors="pt").to(device)
+            image, has_nsfw_concept = self.safety_checker(
+                images=image, clip_input=safety_checker_input.pixel_values.to(dtype)
+            )
+        else:
+            has_nsfw_concept = None
+        return image, has_nsfw_concept
+
+    # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.decode_latents
+    def decode_latents(self, latents):
+        latents = 1 / 0.18215 * latents
+        image = self.vae.decode(latents).sample
+        image = (image / 2 + 0.5).clamp(0, 1)
+        # we always cast to float32 as this does not cause significant overhead and is compatible with bfloa16
+        image = image.cpu().permute(0, 2, 3, 1).float().numpy()
+        return image
+
+    # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion_img2img.StableDiffusionImg2ImgPipeline.get_timesteps
+    def get_timesteps(self, num_inference_steps, strength, device):
+        # get the original timestep using init_timestep
+        offset = self.scheduler.config.get("steps_offset", 0)
+        init_timestep = int(num_inference_steps * strength) + offset
+        init_timestep = min(init_timestep, num_inference_steps)
+
+        t_start = max(num_inference_steps - init_timestep + offset, 0)
+        timesteps = self.scheduler.timesteps[t_start:]
+
+        return timesteps
+
+    def prepare_latents(self, init_image, timestep, batch_size, num_images_per_prompt, dtype, device, generator=None):
+        init_image = init_image.to(device=device, dtype=dtype)
+        init_latent_dist = self.vae.encode(init_image).latent_dist
+        init_latents = init_latent_dist.sample(generator=generator)
+        init_latents = 0.18215 * init_latents
+
+        if batch_size > init_latents.shape[0] and batch_size % init_latents.shape[0] == 0:
+            # expand init_latents for batch_size
+            deprecation_message = (
+                f"You have passed {batch_size} text prompts (`prompt`), but only {init_latents.shape[0]} initial"
+                " images (`init_image`). Initial images are now duplicating to match the number of text prompts. Note"
+                " that this behavior is deprecated and will be removed in a version 1.0.0. Please make sure to update"
+                " your script to pass as many init images as text prompts to suppress this warning."
+            )
+            deprecate("len(prompt) != len(init_image)", "1.0.0", deprecation_message, standard_warn=False)
+            additional_image_per_prompt = batch_size // init_latents.shape[0]
+            init_latents = torch.cat([init_latents] * additional_image_per_prompt * num_images_per_prompt, dim=0)
+        elif batch_size > init_latents.shape[0] and batch_size % init_latents.shape[0] != 0:
+            raise ValueError(
+                f"Cannot duplicate `init_image` of batch size {init_latents.shape[0]} to {batch_size} text prompts."
+            )
+        else:
+            init_latents = torch.cat([init_latents] * num_images_per_prompt, dim=0)
+
+        # add noise to latents using the timestep
+        noise = torch.randn(init_latents.shape, generator=generator, device=device, dtype=dtype)
+
+        # get latents
+        clean_latents = init_latents
+        init_latents = self.scheduler.add_noise(init_latents, noise, timestep)
+        latents = init_latents
+
+        return latents, clean_latents
+
     @torch.no_grad()
     def __call__(
         self,
@@ -384,112 +525,43 @@ class CycleDiffusionPipeline(DiffusionPipeline):
             list of `bool`s denoting whether the corresponding generated image likely represents "not-safe-for-work"
             (nsfw) content, according to the `safety_checker`.
         """
-        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 batch_size != 1:
-            raise ValueError(
-                "At the moment only `batch_size=1` is supported for prompts, but you seem to have passed multiple"
-                f" prompts: {prompt}. Please make sure to pass only a single prompt."
-            )
-
-        if strength < 0 or strength > 1:
-            raise ValueError(f"The value of strength should in [0.0, 1.0] but is {strength}")
-
-        if (callback_steps is None) or (
-            callback_steps is not None and (not isinstance(callback_steps, int) or callback_steps <= 0)
-        ):
-            raise ValueError(
-                f"`callback_steps` has to be a positive integer but is {callback_steps} of type"
-                f" {type(callback_steps)}."
-            )
-
-        # set timesteps
-        self.scheduler.set_timesteps(num_inference_steps)
-
-        if isinstance(init_image, PIL.Image.Image):
-            init_image = preprocess(init_image)
+        # 1. Check inputs
+        self.check_inputs(prompt, strength, callback_steps)
 
+        # 2. Define call parameters
+        batch_size = 1 if isinstance(prompt, str) else len(prompt)
         device = self._execution_device
-
         # 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
 
+        # 3. Encode input prompt
         text_embeddings = self._encode_prompt(prompt, device, num_images_per_prompt, do_classifier_free_guidance, None)
         source_text_embeddings = self._encode_prompt(
             source_prompt, device, num_images_per_prompt, do_classifier_free_guidance, None
         )
 
-        # encode the init image into latents and scale the latents
-        latents_dtype = text_embeddings.dtype
-        init_image = init_image.to(device=self.device, dtype=latents_dtype)
-        init_latent_dist = self.vae.encode(init_image).latent_dist
-        init_latents = init_latent_dist.sample(generator=generator)
-        init_latents = 0.18215 * init_latents
-
-        if isinstance(prompt, str):
-            prompt = [prompt]
-        if len(prompt) > init_latents.shape[0] and len(prompt) % init_latents.shape[0] == 0:
-            # expand init_latents for batch_size
-            deprecation_message = (
-                f"You have passed {len(prompt)} text prompts (`prompt`), but only {init_latents.shape[0]} initial"
-                " images (`init_image`). Initial images are now duplicating to match the number of text prompts. Note"
-                " that this behavior is deprecated and will be removed in a version 1.0.0. Please make sure to update"
-                " your script to pass as many init images as text prompts to suppress this warning."
-            )
-            deprecate("len(prompt) != len(init_image)", "1.0.0", deprecation_message, standard_warn=False)
-            additional_image_per_prompt = len(prompt) // init_latents.shape[0]
-            init_latents = torch.cat([init_latents] * additional_image_per_prompt * num_images_per_prompt, dim=0)
-        elif len(prompt) > init_latents.shape[0] and len(prompt) % init_latents.shape[0] != 0:
-            raise ValueError(
-                f"Cannot duplicate `init_image` of batch size {init_latents.shape[0]} to {len(prompt)} text prompts."
-            )
-        else:
-            init_latents = torch.cat([init_latents] * num_images_per_prompt, dim=0)
-
-        # get the original timestep using init_timestep
-        offset = self.scheduler.config.get("steps_offset", 0)
-        init_timestep = int(num_inference_steps * strength) + offset
-        init_timestep = min(init_timestep, num_inference_steps)
-
-        timesteps = self.scheduler.timesteps[-init_timestep]
-        timesteps = torch.tensor([timesteps] * batch_size * num_images_per_prompt, device=self.device)
-
-        # add noise to latents using the timesteps
-        noise = torch.randn(init_latents.shape, generator=generator, device=self.device, dtype=latents_dtype)
-        clean_latents = init_latents
-        init_latents = self.scheduler.add_noise(init_latents, noise, timesteps)
+        # 4. Preprocess image
+        if isinstance(init_image, PIL.Image.Image):
+            init_image = preprocess(init_image)
 
-        # 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 = {}
+        # 5. Prepare timesteps
+        self.scheduler.set_timesteps(num_inference_steps, device=device)
+        timesteps = self.get_timesteps(num_inference_steps, strength, device)
+        latent_timestep = timesteps[:1].repeat(batch_size * num_images_per_prompt)
 
-        if not (accepts_eta and (0 < eta <= 1)):
-            raise ValueError(
-                "Currently, only the DDIM scheduler is supported. Please make sure that `pipeline.scheduler` is of"
-                f" type {DDIMScheduler.__class__} and not {self.scheduler.__class__}."
+        # 6. Prepare latent variables
+        latents, clean_latents = self.prepare_latents(
+            init_image, latent_timestep, batch_size, num_images_per_prompt, text_embeddings.dtype, device, generator
         )
+        source_latents = latents
 
-        extra_step_kwargs["eta"] = eta
-
-        latents = init_latents
-        source_latents = init_latents
-
-        t_start = max(num_inference_steps - init_timestep + offset, 0)
-
-        # Some schedulers like PNDM have timesteps as arrays
-        # It's more optimized to move all timesteps to correct device beforehand
-        timesteps = self.scheduler.timesteps[t_start:].to(self.device)
+        # 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)
+        generator = extra_step_kwargs.pop("generator", None)
 
+        # 8. Denoising loop
         for i, t in enumerate(self.progress_bar(timesteps)):
             # expand the latents if we are doing classifier free guidance
             latent_model_input = torch.cat([latents] * 2)
@@ -551,22 +623,13 @@ class CycleDiffusionPipeline(DiffusionPipeline):
             if callback is not None and i % callback_steps == 0:
                 callback(i, t, latents)
 
-        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()
+        # 9. Post-processing
+        image = self.decode_latents(latents)
 
-        if self.safety_checker is not None:
-            safety_checker_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_checker_input.pixel_values.to(text_embeddings.dtype)
-            )
-        else:
-            has_nsfw_concept = None
+        # 10. Run safety checker
+        image, has_nsfw_concept = self.run_safety_checker(image, device, text_embeddings.dtype)
 
+        # 11. Convert to PIL
         if output_type == "pil":
             image = self.numpy_to_pil(image)
 

src/diffusers/pipelines/stable_diffusion/pipeline_stable_diffusion.py

@@ -298,6 +298,73 @@ class StableDiffusionPipeline(DiffusionPipeline):
 
         return text_embeddings
 
+    def run_safety_checker(self, image, device, dtype):
+        if self.safety_checker is not None:
+            safety_checker_input = self.feature_extractor(self.numpy_to_pil(image), return_tensors="pt").to(device)
+            image, has_nsfw_concept = self.safety_checker(
+                images=image, clip_input=safety_checker_input.pixel_values.to(dtype)
+            )
+        else:
+            has_nsfw_concept = None
+        return image, has_nsfw_concept
+
+    def decode_latents(self, latents):
+        latents = 1 / 0.18215 * latents
+        image = self.vae.decode(latents).sample
+        image = (image / 2 + 0.5).clamp(0, 1)
+        # we always cast to float32 as this does not cause significant overhead and is compatible with bfloa16
+        image = image.cpu().permute(0, 2, 3, 1).float().numpy()
+        return image
+
+    def prepare_extra_step_kwargs(self, generator, eta):
+        # prepare extra kwargs for the scheduler step, since not all schedulers have the same signature
+        # eta (η) is only used with the DDIMScheduler, it will be ignored for other schedulers.
+        # eta corresponds to η in DDIM paper: https://arxiv.org/abs/2010.02502
+        # and should be between [0, 1]
+
+        accepts_eta = "eta" in set(inspect.signature(self.scheduler.step).parameters.keys())
+        extra_step_kwargs = {}
+        if accepts_eta:
+            extra_step_kwargs["eta"] = eta
+
+        # check if the scheduler accepts generator
+        accepts_generator = "generator" in set(inspect.signature(self.scheduler.step).parameters.keys())
+        if accepts_generator:
+            extra_step_kwargs["generator"] = generator
+        return extra_step_kwargs
+
+    def check_inputs(self, prompt, height, width, callback_steps):
+        if not isinstance(prompt, str) and not isinstance(prompt, list):
+            raise ValueError(f"`prompt` has to be of type `str` or `list` but is {type(prompt)}")
+
+        if height % 8 != 0 or width % 8 != 0:
+            raise ValueError(f"`height` and `width` have to be divisible by 8 but are {height} and {width}.")
+
+        if (callback_steps is None) or (
+            callback_steps is not None and (not isinstance(callback_steps, int) or callback_steps <= 0)
+        ):
+            raise ValueError(
+                f"`callback_steps` has to be a positive integer but is {callback_steps} of type"
+                f" {type(callback_steps)}."
+            )
+
+    def prepare_latents(self, batch_size, num_channels_latents, height, width, dtype, device, generator, latents=None):
+        shape = (batch_size, num_channels_latents, height // 8, width // 8)
+        if latents is None:
+            if device.type == "mps":
+                # randn does not work reproducibly on mps
+                latents = torch.randn(shape, generator=generator, device="cpu", dtype=dtype).to(device)
+            else:
+                latents = torch.randn(shape, generator=generator, device=device, dtype=dtype)
+        else:
+            if latents.shape != shape:
+                raise ValueError(f"Unexpected latents shape, got {latents.shape}, expected {shape}")
+            latents = latents.to(device)
+
+        # scale the initial noise by the standard deviation required by the scheduler
+        latents = latents * self.scheduler.init_noise_sigma
+        return latents
+
     @torch.no_grad()
     def __call__(
         self,
@@ -371,75 +438,45 @@ class StableDiffusionPipeline(DiffusionPipeline):
             list of `bool`s denoting whether the corresponding generated image likely represents "not-safe-for-work"
             (nsfw) content, according to the `safety_checker`.
         """
-        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}.")
-
-        if (callback_steps is None) or (
-            callback_steps is not None and (not isinstance(callback_steps, int) or callback_steps <= 0)
-        ):
-            raise ValueError(
-                f"`callback_steps` has to be a positive integer but is {callback_steps} of type"
-                f" {type(callback_steps)}."
-            )
+        # 1. Check inputs. Raise error if not correct
+        self.check_inputs(prompt, height, width, callback_steps)
 
+        # 2. Define call parameters
+        batch_size = 1 if isinstance(prompt, str) else len(prompt)
         device = self._execution_device
-
         # 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
 
+        # 3. Encode input prompt
         text_embeddings = self._encode_prompt(
             prompt, device, num_images_per_prompt, do_classifier_free_guidance, negative_prompt
         )
 
-        # 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`.
-
-        # get the initial random noise unless the user supplied it
-        latents_shape = (batch_size * num_images_per_prompt, self.unet.in_channels, height // 8, width // 8)
-        latents_dtype = text_embeddings.dtype
-        if latents is None:
-            if device.type == "mps":
-                # randn does not work reproducibly on mps
-                latents = torch.randn(latents_shape, generator=generator, device="cpu", dtype=latents_dtype).to(device)
-            else:
-                latents = torch.randn(latents_shape, generator=generator, device=device, dtype=latents_dtype)
-        else:
-            if latents.shape != latents_shape:
-                raise ValueError(f"Unexpected latents shape, got {latents.shape}, expected {latents_shape}")
-            latents = latents.to(device)
-
-        # set timesteps and move to the correct device
+        # 4. Prepare timesteps
         self.scheduler.set_timesteps(num_inference_steps, device=device)
-        timesteps_tensor = self.scheduler.timesteps
-
-        # scale the initial noise by the standard deviation required by the scheduler
-        latents = latents * self.scheduler.init_noise_sigma
-
-        # 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
+        timesteps = self.scheduler.timesteps
+
+        # 5. Prepare latent variables
+        num_channels_latents = self.unet.in_channels
+        latents = self.prepare_latents(
+            batch_size * num_images_per_prompt,
+            num_channels_latents,
+            height,
+            width,
+            text_embeddings.dtype,
+            device,
+            generator,
+            latents,
+        )
 
-        # check if the scheduler accepts generator
-        accepts_generator = "generator" in set(inspect.signature(self.scheduler.step).parameters.keys())
-        if accepts_generator:
-            extra_step_kwargs["generator"] = generator
+        # 6. 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)
 
-        for i, t in enumerate(self.progress_bar(timesteps_tensor)):
+        # 7. Denoising loop
+        for i, t in enumerate(self.progress_bar(timesteps)):
             # expand the latents if we are doing classifier free guidance
             latent_model_input = torch.cat([latents] * 2) if do_classifier_free_guidance else latents
             latent_model_input = self.scheduler.scale_model_input(latent_model_input, t)
@@ -459,22 +496,13 @@ class StableDiffusionPipeline(DiffusionPipeline):
             if callback is not None and i % callback_steps == 0:
                 callback(i, t, latents)
 
-        latents = 1 / 0.18215 * latents
-        image = self.vae.decode(latents).sample
-
-        image = (image / 2 + 0.5).clamp(0, 1)
-
-        # we always cast to float32 as this does not cause significant overhead and is compatible with bfloa16
-        image = image.cpu().permute(0, 2, 3, 1).float().numpy()
+        # 8. Post-processing
+        image = self.decode_latents(latents)
 
-        if self.safety_checker is not None:
-            safety_checker_input = self.feature_extractor(self.numpy_to_pil(image), return_tensors="pt").to(device)
-            image, has_nsfw_concept = self.safety_checker(
-                images=image, clip_input=safety_checker_input.pixel_values.to(text_embeddings.dtype)
-            )
-        else:
-            has_nsfw_concept = None
+        # 9. Run safety checker
+        image, has_nsfw_concept = self.run_safety_checker(image, device, text_embeddings.dtype)
 
+        # 10. Convert to PIL
         if output_type == "pil":
             image = self.numpy_to_pil(image)
 

src/diffusers/pipelines/stable_diffusion/pipeline_stable_diffusion_img2img.py

@@ -27,6 +27,7 @@ from ...models import AutoencoderKL, UNet2DConditionModel
 from ...pipeline_utils import DiffusionPipeline
 from ...schedulers import (
     DDIMScheduler,
+    DPMSolverMultistepScheduler,
     EulerAncestralDiscreteScheduler,
     EulerDiscreteScheduler,
     LMSDiscreteScheduler,
@@ -78,6 +79,7 @@ class StableDiffusionImg2ImgPipeline(DiffusionPipeline):
             Model that extracts features from generated images to be used as inputs for the `safety_checker`.
     """
 
+    # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.__init__
     def __init__(
         self,
         vae: AutoencoderKL,
@@ -85,7 +87,12 @@ class StableDiffusionImg2ImgPipeline(DiffusionPipeline):
         tokenizer: CLIPTokenizer,
         unet: UNet2DConditionModel,
         scheduler: Union[
-            DDIMScheduler, PNDMScheduler, LMSDiscreteScheduler, EulerDiscreteScheduler, EulerAncestralDiscreteScheduler
+            DDIMScheduler,
+            PNDMScheduler,
+            LMSDiscreteScheduler,
+            EulerDiscreteScheduler,
+            EulerAncestralDiscreteScheduler,
+            DPMSolverMultistepScheduler,
         ],
         safety_checker: StableDiffusionSafetyChecker,
         feature_extractor: CLIPFeatureExtractor,
@@ -139,6 +146,7 @@ class StableDiffusionImg2ImgPipeline(DiffusionPipeline):
             feature_extractor=feature_extractor,
         )
 
+    # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.enable_attention_slicing
     def enable_attention_slicing(self, slice_size: Optional[Union[str, int]] = "auto"):
         r"""
         Enable sliced attention computation.
@@ -158,14 +166,16 @@ class StableDiffusionImg2ImgPipeline(DiffusionPipeline):
             slice_size = self.unet.config.attention_head_dim // 2
         self.unet.set_attention_slice(slice_size)
 
+    # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.disable_attention_slicing
     def disable_attention_slicing(self):
         r"""
         Disable sliced attention computation. If `enable_attention_slicing` was previously invoked, this method will go
         back to computing attention in one step.
         """
-        # set slice_size = `None` to disable `set_attention_slice`
+        # set slice_size = `None` to disable `attention slicing`
         self.enable_attention_slicing(None)
 
+    # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.enable_sequential_cpu_offload
     def enable_sequential_cpu_offload(self):
         r"""
         Offloads all models to CPU using accelerate, significantly reducing memory usage. When called, unet,
@@ -202,6 +212,7 @@ class StableDiffusionImg2ImgPipeline(DiffusionPipeline):
                 return torch.device(module._hf_hook.execution_device)
         return self.device
 
+    # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.enable_xformers_memory_efficient_attention
     def enable_xformers_memory_efficient_attention(self):
         r"""
         Enable memory efficient attention as implemented in xformers.
@@ -214,6 +225,7 @@ class StableDiffusionImg2ImgPipeline(DiffusionPipeline):
         """
         self.unet.set_use_memory_efficient_attention_xformers(True)
 
+    # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.disable_xformers_memory_efficient_attention
     def disable_xformers_memory_efficient_attention(self):
         r"""
         Disable memory efficient attention as implemented in xformers.
@@ -306,6 +318,103 @@ class StableDiffusionImg2ImgPipeline(DiffusionPipeline):
 
         return text_embeddings
 
+    # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.run_safety_checker
+    def run_safety_checker(self, image, device, dtype):
+        if self.safety_checker is not None:
+            safety_checker_input = self.feature_extractor(self.numpy_to_pil(image), return_tensors="pt").to(device)
+            image, has_nsfw_concept = self.safety_checker(
+                images=image, clip_input=safety_checker_input.pixel_values.to(dtype)
+            )
+        else:
+            has_nsfw_concept = None
+        return image, has_nsfw_concept
+
+    # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.decode_latents
+    def decode_latents(self, latents):
+        latents = 1 / 0.18215 * latents
+        image = self.vae.decode(latents).sample
+        image = (image / 2 + 0.5).clamp(0, 1)
+        # we always cast to float32 as this does not cause significant overhead and is compatible with bfloa16
+        image = image.cpu().permute(0, 2, 3, 1).float().numpy()
+        return image
+
+    # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.prepare_extra_step_kwargs
+    def prepare_extra_step_kwargs(self, generator, eta):
+        # prepare extra kwargs for the scheduler step, since not all schedulers have the same signature
+        # eta (η) is only used with the DDIMScheduler, it will be ignored for other schedulers.
+        # eta corresponds to η in DDIM paper: https://arxiv.org/abs/2010.02502
+        # and should be between [0, 1]
+
+        accepts_eta = "eta" in set(inspect.signature(self.scheduler.step).parameters.keys())
+        extra_step_kwargs = {}
+        if accepts_eta:
+            extra_step_kwargs["eta"] = eta
+
+        # check if the scheduler accepts generator
+        accepts_generator = "generator" in set(inspect.signature(self.scheduler.step).parameters.keys())
+        if accepts_generator:
+            extra_step_kwargs["generator"] = generator
+        return extra_step_kwargs
+
+    def check_inputs(self, prompt, strength, callback_steps):
+        if not isinstance(prompt, str) and not isinstance(prompt, list):
+            raise ValueError(f"`prompt` has to be of type `str` or `list` but is {type(prompt)}")
+
+        if strength < 0 or strength > 1:
+            raise ValueError(f"The value of strength should in [1.0, 1.0] but is {strength}")
+
+        if (callback_steps is None) or (
+            callback_steps is not None and (not isinstance(callback_steps, int) or callback_steps <= 0)
+        ):
+            raise ValueError(
+                f"`callback_steps` has to be a positive integer but is {callback_steps} of type"
+                f" {type(callback_steps)}."
+            )
+
+    def get_timesteps(self, num_inference_steps, strength, device):
+        # get the original timestep using init_timestep
+        offset = self.scheduler.config.get("steps_offset", 0)
+        init_timestep = int(num_inference_steps * strength) + offset
+        init_timestep = min(init_timestep, num_inference_steps)
+
+        t_start = max(num_inference_steps - init_timestep + offset, 0)
+        timesteps = self.scheduler.timesteps[t_start:]
+
+        return timesteps
+
+    def prepare_latents(self, init_image, timestep, batch_size, num_images_per_prompt, dtype, device, generator=None):
+        init_image = init_image.to(device=device, dtype=dtype)
+        init_latent_dist = self.vae.encode(init_image).latent_dist
+        init_latents = init_latent_dist.sample(generator=generator)
+        init_latents = 0.18215 * init_latents
+
+        if batch_size > init_latents.shape[0] and batch_size % init_latents.shape[0] == 0:
+            # expand init_latents for batch_size
+            deprecation_message = (
+                f"You have passed {batch_size} text prompts (`prompt`), but only {init_latents.shape[0]} initial"
+                " images (`init_image`). Initial images are now duplicating to match the number of text prompts. Note"
+                " that this behavior is deprecated and will be removed in a version 1.0.0. Please make sure to update"
+                " your script to pass as many init images as text prompts to suppress this warning."
+            )
+            deprecate("len(prompt) != len(init_image)", "1.0.0", deprecation_message, standard_warn=False)
+            additional_image_per_prompt = batch_size // init_latents.shape[0]
+            init_latents = torch.cat([init_latents] * additional_image_per_prompt * num_images_per_prompt, dim=0)
+        elif batch_size > init_latents.shape[0] and batch_size % init_latents.shape[0] != 0:
+            raise ValueError(
+                f"Cannot duplicate `init_image` of batch size {init_latents.shape[0]} to {batch_size} text prompts."
+            )
+        else:
+            init_latents = torch.cat([init_latents] * num_images_per_prompt, dim=0)
+
+        # add noise to latents using the timesteps
+        noise = torch.randn(init_latents.shape, generator=generator, device=device, dtype=dtype)
+
+        # get latents
+        init_latents = self.scheduler.add_noise(init_latents, noise, timestep)
+        latents = init_latents
+
+        return latents
+
     @torch.no_grad()
     def __call__(
         self,
@@ -379,102 +488,40 @@ class StableDiffusionImg2ImgPipeline(DiffusionPipeline):
             list of `bool`s denoting whether the corresponding generated image likely represents "not-safe-for-work"
             (nsfw) content, according to the `safety_checker`.
         """
-        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 strength < 0 or strength > 1:
-            raise ValueError(f"The value of strength should in [0.0, 1.0] but is {strength}")
-
-        if (callback_steps is None) or (
-            callback_steps is not None and (not isinstance(callback_steps, int) or callback_steps <= 0)
-        ):
-            raise ValueError(
-                f"`callback_steps` has to be a positive integer but is {callback_steps} of type"
-                f" {type(callback_steps)}."
-            )
+        # 1. Check inputs
+        self.check_inputs(prompt, strength, callback_steps)
 
+        # 2. Define call parameters
+        batch_size = 1 if isinstance(prompt, str) else len(prompt)
         device = self._execution_device
-
-        # set timesteps
-        self.scheduler.set_timesteps(num_inference_steps)
-
-        if isinstance(init_image, PIL.Image.Image):
-            init_image = preprocess(init_image)
-
         # 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
 
+        # 3. Encode input prompt
         text_embeddings = self._encode_prompt(
             prompt, device, num_images_per_prompt, do_classifier_free_guidance, negative_prompt
         )
 
-        # encode the init image into latents and scale the latents
-        latents_dtype = text_embeddings.dtype
-        init_image = init_image.to(device=device, dtype=latents_dtype)
-        init_latent_dist = self.vae.encode(init_image).latent_dist
-        init_latents = init_latent_dist.sample(generator=generator)
-        init_latents = 0.18215 * init_latents
-
-        if isinstance(prompt, str):
-            prompt = [prompt]
-        if len(prompt) > init_latents.shape[0] and len(prompt) % init_latents.shape[0] == 0:
-            # expand init_latents for batch_size
-            deprecation_message = (
-                f"You have passed {len(prompt)} text prompts (`prompt`), but only {init_latents.shape[0]} initial"
-                " images (`init_image`). Initial images are now duplicating to match the number of text prompts. Note"
-                " that this behavior is deprecated and will be removed in a version 1.0.0. Please make sure to update"
-                " your script to pass as many init images as text prompts to suppress this warning."
-            )
-            deprecate("len(prompt) != len(init_image)", "1.0.0", deprecation_message, standard_warn=False)
-            additional_image_per_prompt = len(prompt) // init_latents.shape[0]
-            init_latents = torch.cat([init_latents] * additional_image_per_prompt * num_images_per_prompt, dim=0)
-        elif len(prompt) > init_latents.shape[0] and len(prompt) % init_latents.shape[0] != 0:
-            raise ValueError(
-                f"Cannot duplicate `init_image` of batch size {init_latents.shape[0]} to {len(prompt)} text prompts."
-            )
-        else:
-            init_latents = torch.cat([init_latents] * num_images_per_prompt, dim=0)
-
-        # get the original timestep using init_timestep
-        offset = self.scheduler.config.get("steps_offset", 0)
-        init_timestep = int(num_inference_steps * strength) + offset
-        init_timestep = min(init_timestep, num_inference_steps)
-
-        timesteps = self.scheduler.timesteps[-init_timestep]
-        timesteps = torch.tensor([timesteps] * batch_size * num_images_per_prompt, device=device)
-
-        # add noise to latents using the timesteps
-        noise = torch.randn(init_latents.shape, generator=generator, device=device, dtype=latents_dtype)
-        init_latents = self.scheduler.add_noise(init_latents, noise, timesteps)
-
-        # prepare extra kwargs for the scheduler step, since not all schedulers have the same signature
-        # eta (η) is only used with the DDIMScheduler, it will be ignored for other schedulers.
-        # eta corresponds to η in DDIM paper: https://arxiv.org/abs/2010.02502
-        # and should be between [0, 1]
-        accepts_eta = "eta" in set(inspect.signature(self.scheduler.step).parameters.keys())
-        extra_step_kwargs = {}
-        if accepts_eta:
-            extra_step_kwargs["eta"] = eta
-
-        # check if the scheduler accepts generator
-        accepts_generator = "generator" in set(inspect.signature(self.scheduler.step).parameters.keys())
-        if accepts_generator:
-            extra_step_kwargs["generator"] = generator
+        # 4. Preprocess image
+        if isinstance(init_image, PIL.Image.Image):
+            init_image = preprocess(init_image)
 
-        latents = init_latents
+        # 5. set timesteps
+        self.scheduler.set_timesteps(num_inference_steps, device=device)
+        timesteps = self.get_timesteps(num_inference_steps, strength, device)
+        latent_timestep = timesteps[:1].repeat(batch_size * num_images_per_prompt)
 
-        t_start = max(num_inference_steps - init_timestep + offset, 0)
+        # 6. Prepare latent variables
+        latents = self.prepare_latents(
+            init_image, latent_timestep, batch_size, num_images_per_prompt, text_embeddings.dtype, device, generator
+        )
 
-        # Some schedulers like PNDM have timesteps as arrays
-        # It's more optimized to move all timesteps to correct device beforehand
-        timesteps = self.scheduler.timesteps[t_start:].to(device)
+        # 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)
 
+        # 8. Denoising loop
         for i, t in enumerate(self.progress_bar(timesteps)):
             # expand the latents if we are doing classifier free guidance
             latent_model_input = torch.cat([latents] * 2) if do_classifier_free_guidance else latents
@@ -495,20 +542,13 @@ class StableDiffusionImg2ImgPipeline(DiffusionPipeline):
             if callback is not None and i % callback_steps == 0:
                 callback(i, t, latents)
 
-        latents = 1 / 0.18215 * latents
-        image = self.vae.decode(latents).sample
+        # 9. Post-processing
+        image = self.decode_latents(latents)
 
-        image = (image / 2 + 0.5).clamp(0, 1)
-        image = image.cpu().permute(0, 2, 3, 1).numpy()
-
-        if self.safety_checker is not None:
-            safety_checker_input = self.feature_extractor(self.numpy_to_pil(image), return_tensors="pt").to(device)
-            image, has_nsfw_concept = self.safety_checker(
-                images=image, clip_input=safety_checker_input.pixel_values.to(text_embeddings.dtype)
-            )
-        else:
-            has_nsfw_concept = None
+        # 10. Run safety checker
+        image, has_nsfw_concept = self.run_safety_checker(image, device, text_embeddings.dtype)
 
+        # 11. Convert to PIL
         if output_type == "pil":
             image = self.numpy_to_pil(image)
 

src/diffusers/pipelines/stable_diffusion/pipeline_stable_diffusion_inpaint.py

@@ -139,6 +139,7 @@ class StableDiffusionInpaintPipeline(DiffusionPipeline):
             feature_extractor=feature_extractor,
         )
 
+    # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.enable_attention_slicing
     def enable_attention_slicing(self, slice_size: Optional[Union[str, int]] = "auto"):
         r"""
         Enable sliced attention computation.
@@ -158,6 +159,7 @@ class StableDiffusionInpaintPipeline(DiffusionPipeline):
             slice_size = self.unet.config.attention_head_dim // 2
         self.unet.set_attention_slice(slice_size)
 
+    # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.disable_attention_slicing
     def disable_attention_slicing(self):
         r"""
         Disable sliced attention computation. If `enable_attention_slicing` was previously invoked, this method will go
@@ -166,6 +168,7 @@ class StableDiffusionInpaintPipeline(DiffusionPipeline):
         # set slice_size = `None` to disable `attention slicing`
         self.enable_attention_slicing(None)
 
+    # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.enable_sequential_cpu_offload
     def enable_sequential_cpu_offload(self):
         r"""
         Offloads all models to CPU using accelerate, significantly reducing memory usage. When called, unet,
@@ -183,6 +186,26 @@ class StableDiffusionInpaintPipeline(DiffusionPipeline):
             if cpu_offloaded_model is not None:
                 cpu_offload(cpu_offloaded_model, device)
 
+    # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.enable_xformers_memory_efficient_attention
+    def enable_xformers_memory_efficient_attention(self):
+        r"""
+        Enable memory efficient attention as implemented in xformers.
+
+        When this option is enabled, you should observe lower GPU memory usage and a potential speed up at inference
+        time. Speed up at training time is not guaranteed.
+
+        Warning: When Memory Efficient Attention and Sliced attention are both enabled, the Memory Efficient Attention
+        is used.
+        """
+        self.unet.set_use_memory_efficient_attention_xformers(True)
+
+    # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.disable_xformers_memory_efficient_attention
+    def disable_xformers_memory_efficient_attention(self):
+        r"""
+        Disable memory efficient attention as implemented in xformers.
+        """
+        self.unet.set_use_memory_efficient_attention_xformers(False)
+
     @property
     # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline._execution_device
     def _execution_device(self):
@@ -202,24 +225,6 @@ class StableDiffusionInpaintPipeline(DiffusionPipeline):
                 return torch.device(module._hf_hook.execution_device)
         return self.device
 
-    def enable_xformers_memory_efficient_attention(self):
-        r"""
-        Enable memory efficient attention as implemented in xformers.
-
-        When this option is enabled, you should observe lower GPU memory usage and a potential speed up at inference
-        time. Speed up at training time is not guaranteed.
-
-        Warning: When Memory Efficient Attention and Sliced attention are both enabled, the Memory Efficient Attention
-        is used.
-        """
-        self.unet.set_use_memory_efficient_attention_xformers(True)
-
-    def disable_xformers_memory_efficient_attention(self):
-        r"""
-        Disable memory efficient attention as implemented in xformers.
-        """
-        self.unet.set_use_memory_efficient_attention_xformers(False)
-
     # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline._encode_prompt
     def _encode_prompt(self, prompt, device, num_images_per_prompt, do_classifier_free_guidance, negative_prompt):
         r"""
@@ -306,6 +311,106 @@ class StableDiffusionInpaintPipeline(DiffusionPipeline):
 
         return text_embeddings
 
+    # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.run_safety_checker
+    def run_safety_checker(self, image, device, dtype):
+        if self.safety_checker is not None:
+            safety_checker_input = self.feature_extractor(self.numpy_to_pil(image), return_tensors="pt").to(device)
+            image, has_nsfw_concept = self.safety_checker(
+                images=image, clip_input=safety_checker_input.pixel_values.to(dtype)
+            )
+        else:
+            has_nsfw_concept = None
+        return image, has_nsfw_concept
+
+    # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.prepare_extra_step_kwargs
+    def prepare_extra_step_kwargs(self, generator, eta):
+        # prepare extra kwargs for the scheduler step, since not all schedulers have the same signature
+        # eta (η) is only used with the DDIMScheduler, it will be ignored for other schedulers.
+        # eta corresponds to η in DDIM paper: https://arxiv.org/abs/2010.02502
+        # and should be between [0, 1]
+
+        accepts_eta = "eta" in set(inspect.signature(self.scheduler.step).parameters.keys())
+        extra_step_kwargs = {}
+        if accepts_eta:
+            extra_step_kwargs["eta"] = eta
+
+        # check if the scheduler accepts generator
+        accepts_generator = "generator" in set(inspect.signature(self.scheduler.step).parameters.keys())
+        if accepts_generator:
+            extra_step_kwargs["generator"] = generator
+        return extra_step_kwargs
+
+    # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.decode_latents
+    def decode_latents(self, latents):
+        latents = 1 / 0.18215 * latents
+        image = self.vae.decode(latents).sample
+        image = (image / 2 + 0.5).clamp(0, 1)
+        # we always cast to float32 as this does not cause significant overhead and is compatible with bfloa16
+        image = image.cpu().permute(0, 2, 3, 1).float().numpy()
+        return image
+
+    # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.check_inputs
+    def check_inputs(self, prompt, height, width, callback_steps):
+        if not isinstance(prompt, str) and not isinstance(prompt, list):
+            raise ValueError(f"`prompt` has to be of type `str` or `list` but is {type(prompt)}")
+
+        if height % 8 != 0 or width % 8 != 0:
+            raise ValueError(f"`height` and `width` have to be divisible by 8 but are {height} and {width}.")
+
+        if (callback_steps is None) or (
+            callback_steps is not None and (not isinstance(callback_steps, int) or callback_steps <= 0)
+        ):
+            raise ValueError(
+                f"`callback_steps` has to be a positive integer but is {callback_steps} of type"
+                f" {type(callback_steps)}."
+            )
+
+    # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.prepare_latents
+    def prepare_latents(self, batch_size, num_channels_latents, height, width, dtype, device, generator, latents=None):
+        shape = (batch_size, num_channels_latents, height // 8, width // 8)
+        if latents is None:
+            if device.type == "mps":
+                # randn does not work reproducibly on mps
+                latents = torch.randn(shape, generator=generator, device="cpu", dtype=dtype).to(device)
+            else:
+                latents = torch.randn(shape, generator=generator, device=device, dtype=dtype)
+        else:
+            if latents.shape != shape:
+                raise ValueError(f"Unexpected latents shape, got {latents.shape}, expected {shape}")
+            latents = latents.to(device)
+
+        # scale the initial noise by the standard deviation required by the scheduler
+        latents = latents * self.scheduler.init_noise_sigma
+        return latents
+
+    def prepare_mask_latents(
+        self, mask, masked_image, batch_size, height, width, dtype, device, generator, do_classifier_free_guidance
+    ):
+        # resize the mask to latents shape as we concatenate the mask to the latents
+        # we do that before converting to dtype to avoid breaking in case we're using cpu_offload
+        # and half precision
+        mask = torch.nn.functional.interpolate(mask, size=(height // 8, width // 8))
+        mask = mask.to(device=device, dtype=dtype)
+
+        masked_image = masked_image.to(device=device, dtype=dtype)
+
+        # encode the mask image into latents space so we can concatenate it to the latents
+        masked_image_latents = self.vae.encode(masked_image).latent_dist.sample(generator=generator)
+        masked_image_latents = 0.18215 * masked_image_latents
+
+        # duplicate mask and masked_image_latents for each generation per prompt, using mps friendly method
+        mask = mask.repeat(batch_size, 1, 1, 1)
+        masked_image_latents = masked_image_latents.repeat(batch_size, 1, 1, 1)
+
+        mask = torch.cat([mask] * 2) if do_classifier_free_guidance else mask
+        masked_image_latents = (
+            torch.cat([masked_image_latents] * 2) if do_classifier_free_guidance else masked_image_latents
+        )
+
+        # aligning device to prevent device errors when concating it with the latent model input
+        masked_image_latents = masked_image_latents.to(device=device, dtype=dtype)
+        return mask, masked_image_latents
+
     @torch.no_grad()
     def __call__(
         self,
@@ -390,83 +495,59 @@ class StableDiffusionInpaintPipeline(DiffusionPipeline):
             (nsfw) content, according to the `safety_checker`.
         """
 
-        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}.")
-
-        if (callback_steps is None) or (
-            callback_steps is not None and (not isinstance(callback_steps, int) or callback_steps <= 0)
-        ):
-            raise ValueError(
-                f"`callback_steps` has to be a positive integer but is {callback_steps} of type"
-                f" {type(callback_steps)}."
-            )
+        # 1. Check inputs
+        self.check_inputs(prompt, height, width, callback_steps)
 
+        # 2. Define call parameters
+        batch_size = 1 if isinstance(prompt, str) else len(prompt)
         device = self._execution_device
-
         # 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
 
+        # 3. Encode input prompt
         text_embeddings = self._encode_prompt(
             prompt, device, num_images_per_prompt, do_classifier_free_guidance, negative_prompt
         )
 
-        # 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`.
-        num_channels_latents = self.vae.config.latent_channels
-        latents_shape = (batch_size * num_images_per_prompt, num_channels_latents, height // 8, width // 8)
-        latents_dtype = text_embeddings.dtype
-        if latents is None:
-            if device.type == "mps":
-                # randn does not exist on mps
-                latents = torch.randn(latents_shape, generator=generator, device="cpu", dtype=latents_dtype).to(device)
-            else:
-                latents = torch.randn(latents_shape, generator=generator, device=device, dtype=latents_dtype)
-        else:
-            if latents.shape != latents_shape:
-                raise ValueError(f"Unexpected latents shape, got {latents.shape}, expected {latents_shape}")
-            latents = latents.to(device)
-
-        # prepare mask and masked_image
+        # 4. Preprocess mask and image
+        if isinstance(image, PIL.Image.Image) and isinstance(mask_image, PIL.Image.Image):
             mask, masked_image = prepare_mask_and_masked_image(image, mask_image)
 
-        # resize the mask to latents shape as we concatenate the mask to the latents
-        # we do that before converting to dtype to avoid breaking in case we're using cpu_offload
-        # and half precision
-        mask = torch.nn.functional.interpolate(mask, size=(height // 8, width // 8))
-        mask = mask.to(device=device, dtype=text_embeddings.dtype)
-
-        masked_image = masked_image.to(device=device, dtype=text_embeddings.dtype)
-
-        # encode the mask image into latents space so we can concatenate it to the latents
-        masked_image_latents = self.vae.encode(masked_image).latent_dist.sample(generator=generator)
-        masked_image_latents = 0.18215 * masked_image_latents
-
-        # duplicate mask and masked_image_latents for each generation per prompt, using mps friendly method
-        mask = mask.repeat(batch_size * num_images_per_prompt, 1, 1, 1)
-        masked_image_latents = masked_image_latents.repeat(batch_size * num_images_per_prompt, 1, 1, 1)
+        # 5. set timesteps
+        self.scheduler.set_timesteps(num_inference_steps, device=device)
+        timesteps_tensor = self.scheduler.timesteps
 
-        mask = torch.cat([mask] * 2) if do_classifier_free_guidance else mask
-        masked_image_latents = (
-            torch.cat([masked_image_latents] * 2) if do_classifier_free_guidance else masked_image_latents
+        # 6. Prepare latent variables
+        num_channels_latents = self.vae.config.latent_channels
+        latents = self.prepare_latents(
+            batch_size * num_images_per_prompt,
+            num_channels_latents,
+            height,
+            width,
+            text_embeddings.dtype,
+            device,
+            generator,
+            latents,
         )
 
-        # aligning device to prevent device errors when concating it with the latent model input
-        masked_image_latents = masked_image_latents.to(device=device, dtype=text_embeddings.dtype)
+        # 7. Prepare mask latent variables
+        mask, masked_image_latents = self.prepare_mask_latents(
+            mask,
+            masked_image,
+            batch_size,
+            height,
+            width,
+            text_embeddings.dtype,
+            device,
+            generator,
+            do_classifier_free_guidance,
+        )
 
+        # 8. Check that sizes of mask, masked image and latents match
         num_channels_mask = mask.shape[1]
         num_channels_masked_image = masked_image_latents.shape[1]
-
         if num_channels_latents + num_channels_mask + num_channels_masked_image != self.unet.config.in_channels:
             raise ValueError(
                 f"Incorrect configuration settings! The config of `pipeline.unet`: {self.unet.config} expects"
@@ -476,27 +557,10 @@ class StableDiffusionInpaintPipeline(DiffusionPipeline):
                 " `pipeline.unet` or your `mask_image` or `image` input."
             )
 
-        # set timesteps and move to the correct device
-        self.scheduler.set_timesteps(num_inference_steps, device=device)
-        timesteps_tensor = self.scheduler.timesteps
-
-        # scale the initial noise by the standard deviation required by the scheduler
-        latents = latents * self.scheduler.init_noise_sigma
-
-        # prepare extra kwargs for the scheduler step, since not all schedulers have the same signature
-        # eta (η) is only used with the DDIMScheduler, it will be ignored for other schedulers.
-        # eta corresponds to η in DDIM paper: https://arxiv.org/abs/2010.02502
-        # and should be between [0, 1]
-        accepts_eta = "eta" in set(inspect.signature(self.scheduler.step).parameters.keys())
-        extra_step_kwargs = {}
-        if accepts_eta:
-            extra_step_kwargs["eta"] = eta
-
-        # check if the scheduler accepts generator
-        accepts_generator = "generator" in set(inspect.signature(self.scheduler.step).parameters.keys())
-        if accepts_generator:
-            extra_step_kwargs["generator"] = generator
+        # 9. 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)
 
+        # 10. Denoising loop
         for i, t in enumerate(self.progress_bar(timesteps_tensor)):
             # expand the latents if we are doing classifier free guidance
             latent_model_input = torch.cat([latents] * 2) if do_classifier_free_guidance else latents
@@ -521,22 +585,13 @@ class StableDiffusionInpaintPipeline(DiffusionPipeline):
             if callback is not None and i % callback_steps == 0:
                 callback(i, t, latents)
 
-        latents = 1 / 0.18215 * latents
-        image = self.vae.decode(latents).sample
-
-        image = (image / 2 + 0.5).clamp(0, 1)
-
-        # we always cast to float32 as this does not cause significant overhead and is compatible with bfloa16
-        image = image.cpu().permute(0, 2, 3, 1).float().numpy()
+        # 11. Post-processing
+        image = self.decode_latents(latents)
 
-        if self.safety_checker is not None:
-            safety_checker_input = self.feature_extractor(self.numpy_to_pil(image), return_tensors="pt").to(device)
-            image, has_nsfw_concept = self.safety_checker(
-                images=image, clip_input=safety_checker_input.pixel_values.to(text_embeddings.dtype)
-            )
-        else:
-            has_nsfw_concept = None
+        # 12. Run safety checker
+        image, has_nsfw_concept = self.run_safety_checker(image, device, text_embeddings.dtype)
 
+        # 13. Convert to PIL
         if output_type == "pil":
             image = self.numpy_to_pil(image)
 

src/diffusers/pipelines/stable_diffusion/pipeline_stable_diffusion_inpaint_legacy.py

@@ -19,13 +19,20 @@ import numpy as np
 import torch
 
 import PIL
-from tqdm.auto import tqdm
+from diffusers.utils import is_accelerate_available
 from transformers import CLIPFeatureExtractor, CLIPTextModel, CLIPTokenizer
 
 from ...configuration_utils import FrozenDict
 from ...models import AutoencoderKL, UNet2DConditionModel
 from ...pipeline_utils import DiffusionPipeline
-from ...schedulers import DDIMScheduler, LMSDiscreteScheduler, PNDMScheduler
+from ...schedulers import (
+    DDIMScheduler,
+    DPMSolverMultistepScheduler,
+    EulerAncestralDiscreteScheduler,
+    EulerDiscreteScheduler,
+    LMSDiscreteScheduler,
+    PNDMScheduler,
+)
 from ...utils import deprecate, logging
 from . import StableDiffusionPipelineOutput
 from .safety_checker import StableDiffusionSafetyChecker
@@ -85,17 +92,26 @@ class StableDiffusionInpaintPipelineLegacy(DiffusionPipeline):
             Model that extracts features from generated images to be used as inputs for the `safety_checker`.
     """
 
+    # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.__init__
     def __init__(
         self,
         vae: AutoencoderKL,
         text_encoder: CLIPTextModel,
         tokenizer: CLIPTokenizer,
         unet: UNet2DConditionModel,
-        scheduler: Union[DDIMScheduler, PNDMScheduler, LMSDiscreteScheduler],
+        scheduler: Union[
+            DDIMScheduler,
+            PNDMScheduler,
+            LMSDiscreteScheduler,
+            EulerDiscreteScheduler,
+            EulerAncestralDiscreteScheduler,
+            DPMSolverMultistepScheduler,
+        ],
         safety_checker: StableDiffusionSafetyChecker,
         feature_extractor: CLIPFeatureExtractor,
     ):
         super().__init__()
+
         if hasattr(scheduler.config, "steps_offset") and scheduler.config.steps_offset != 1:
             deprecation_message = (
                 f"The configuration file of this scheduler: {scheduler} is outdated. `steps_offset`"
@@ -143,6 +159,7 @@ class StableDiffusionInpaintPipelineLegacy(DiffusionPipeline):
             feature_extractor=feature_extractor,
         )
 
+    # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.enable_attention_slicing
     def enable_attention_slicing(self, slice_size: Optional[Union[str, int]] = "auto"):
         r"""
         Enable sliced attention computation.
@@ -162,14 +179,53 @@ class StableDiffusionInpaintPipelineLegacy(DiffusionPipeline):
             slice_size = self.unet.config.attention_head_dim // 2
         self.unet.set_attention_slice(slice_size)
 
+    # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.disable_attention_slicing
     def disable_attention_slicing(self):
         r"""
         Disable sliced attention computation. If `enable_attention_slicing` was previously invoked, this method will go
         back to computing attention in one step.
         """
-        # set slice_size = `None` to disable `set_attention_slice`
+        # set slice_size = `None` to disable `attention slicing`
         self.enable_attention_slicing(None)
 
+    # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.enable_sequential_cpu_offload
+    def enable_sequential_cpu_offload(self):
+        r"""
+        Offloads all models to CPU using accelerate, significantly reducing memory usage. When called, unet,
+        text_encoder, vae and safety checker have their state dicts saved to CPU and then are moved to a
+        `torch.device('meta') and loaded to GPU only when their specific submodule has its `forward` method called.
+        """
+        if is_accelerate_available():
+            from accelerate import cpu_offload
+        else:
+            raise ImportError("Please install accelerate via `pip install accelerate`")
+
+        device = torch.device("cuda")
+
+        for cpu_offloaded_model in [self.unet, self.text_encoder, self.vae, self.safety_checker]:
+            if cpu_offloaded_model is not None:
+                cpu_offload(cpu_offloaded_model, device)
+
+    # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.enable_xformers_memory_efficient_attention
+    def enable_xformers_memory_efficient_attention(self):
+        r"""
+        Enable memory efficient attention as implemented in xformers.
+
+        When this option is enabled, you should observe lower GPU memory usage and a potential speed up at inference
+        time. Speed up at training time is not guaranteed.
+
+        Warning: When Memory Efficient Attention and Sliced attention are both enabled, the Memory Efficient Attention
+        is used.
+        """
+        self.unet.set_use_memory_efficient_attention_xformers(True)
+
+    # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.disable_xformers_memory_efficient_attention
+    def disable_xformers_memory_efficient_attention(self):
+        r"""
+        Disable memory efficient attention as implemented in xformers.
+        """
+        self.unet.set_use_memory_efficient_attention_xformers(False)
+
     @property
     # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline._execution_device
     def _execution_device(self):
@@ -275,6 +331,88 @@ class StableDiffusionInpaintPipelineLegacy(DiffusionPipeline):
 
         return text_embeddings
 
+    # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.run_safety_checker
+    def run_safety_checker(self, image, device, dtype):
+        if self.safety_checker is not None:
+            safety_checker_input = self.feature_extractor(self.numpy_to_pil(image), return_tensors="pt").to(device)
+            image, has_nsfw_concept = self.safety_checker(
+                images=image, clip_input=safety_checker_input.pixel_values.to(dtype)
+            )
+        else:
+            has_nsfw_concept = None
+        return image, has_nsfw_concept
+
+    # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.decode_latents
+    def decode_latents(self, latents):
+        latents = 1 / 0.18215 * latents
+        image = self.vae.decode(latents).sample
+        image = (image / 2 + 0.5).clamp(0, 1)
+        # we always cast to float32 as this does not cause significant overhead and is compatible with bfloa16
+        image = image.cpu().permute(0, 2, 3, 1).float().numpy()
+        return image
+
+    # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.prepare_extra_step_kwargs
+    def prepare_extra_step_kwargs(self, generator, eta):
+        # prepare extra kwargs for the scheduler step, since not all schedulers have the same signature
+        # eta (η) is only used with the DDIMScheduler, it will be ignored for other schedulers.
+        # eta corresponds to η in DDIM paper: https://arxiv.org/abs/2010.02502
+        # and should be between [0, 1]
+
+        accepts_eta = "eta" in set(inspect.signature(self.scheduler.step).parameters.keys())
+        extra_step_kwargs = {}
+        if accepts_eta:
+            extra_step_kwargs["eta"] = eta
+
+        # check if the scheduler accepts generator
+        accepts_generator = "generator" in set(inspect.signature(self.scheduler.step).parameters.keys())
+        if accepts_generator:
+            extra_step_kwargs["generator"] = generator
+        return extra_step_kwargs
+
+    # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion_img2img.StableDiffusionImg2ImgPipeline.check_inputs
+    def check_inputs(self, prompt, strength, callback_steps):
+        if not isinstance(prompt, str) and not isinstance(prompt, list):
+            raise ValueError(f"`prompt` has to be of type `str` or `list` but is {type(prompt)}")
+
+        if strength < 0 or strength > 1:
+            raise ValueError(f"The value of strength should in [1.0, 1.0] but is {strength}")
+
+        if (callback_steps is None) or (
+            callback_steps is not None and (not isinstance(callback_steps, int) or callback_steps <= 0)
+        ):
+            raise ValueError(
+                f"`callback_steps` has to be a positive integer but is {callback_steps} of type"
+                f" {type(callback_steps)}."
+            )
+
+    # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion_img2img.StableDiffusionImg2ImgPipeline.get_timesteps
+    def get_timesteps(self, num_inference_steps, strength, device):
+        # get the original timestep using init_timestep
+        offset = self.scheduler.config.get("steps_offset", 0)
+        init_timestep = int(num_inference_steps * strength) + offset
+        init_timestep = min(init_timestep, num_inference_steps)
+
+        t_start = max(num_inference_steps - init_timestep + offset, 0)
+        timesteps = self.scheduler.timesteps[t_start:]
+
+        return timesteps
+
+    def prepare_latents(self, init_image, timestep, batch_size, num_images_per_prompt, dtype, device, generator):
+        init_image = init_image.to(device=self.device, dtype=dtype)
+        init_latent_dist = self.vae.encode(init_image).latent_dist
+        init_latents = init_latent_dist.sample(generator=generator)
+        init_latents = 0.18215 * init_latents
+
+        # Expand init_latents for batch_size and num_images_per_prompt
+        init_latents = torch.cat([init_latents] * batch_size * num_images_per_prompt, dim=0)
+        init_latents_orig = init_latents
+
+        # add noise to latents using the timesteps
+        noise = torch.randn(init_latents.shape, generator=generator, device=self.device, dtype=dtype)
+        init_latents = self.scheduler.add_noise(init_latents, noise, timestep)
+        latents = init_latents
+        return latents, init_latents_orig, noise
+
     @torch.no_grad()
     def __call__(
         self,
@@ -353,98 +491,49 @@ class StableDiffusionInpaintPipelineLegacy(DiffusionPipeline):
             list of `bool`s denoting whether the corresponding generated image likely represents "not-safe-for-work"
             (nsfw) content, according to the `safety_checker`.
         """
-        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 strength < 0 or strength > 1:
-            raise ValueError(f"The value of strength should in [0.0, 1.0] but is {strength}")
-
-        if (callback_steps is None) or (
-            callback_steps is not None and (not isinstance(callback_steps, int) or callback_steps <= 0)
-        ):
-            raise ValueError(
-                f"`callback_steps` has to be a positive integer but is {callback_steps} of type"
-                f" {type(callback_steps)}."
-            )
+        # 1. Check inputs
+        self.check_inputs(prompt, strength, callback_steps)
 
+        # 2. Define call parameters
+        batch_size = 1 if isinstance(prompt, str) else len(prompt)
         device = self._execution_device
-
-        # set timesteps
-        self.scheduler.set_timesteps(num_inference_steps)
-
-        # preprocess image
-        if not isinstance(init_image, torch.FloatTensor):
-            init_image = preprocess_image(init_image)
-
         # 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
 
+        # 3. Encode input prompt
         text_embeddings = self._encode_prompt(
             prompt, device, num_images_per_prompt, do_classifier_free_guidance, negative_prompt
         )
 
-        # encode the init image into latents and scale the latents
-        latents_dtype = text_embeddings.dtype
-        init_image = init_image.to(device=self.device, dtype=latents_dtype)
-        init_latent_dist = self.vae.encode(init_image).latent_dist
-        init_latents = init_latent_dist.sample(generator=generator)
-        init_latents = 0.18215 * init_latents
-
-        # Expand init_latents for batch_size and num_images_per_prompt
-        init_latents = torch.cat([init_latents] * batch_size * num_images_per_prompt, dim=0)
-        init_latents_orig = init_latents
+        # 4. Preprocess image and mask
+        if not isinstance(init_image, torch.FloatTensor):
+            init_image = preprocess_image(init_image)
 
-        # preprocess mask
         if not isinstance(mask_image, torch.FloatTensor):
             mask_image = preprocess_mask(mask_image)
-        mask_image = mask_image.to(device=self.device, dtype=latents_dtype)
-        mask = torch.cat([mask_image] * batch_size * num_images_per_prompt)
 
-        # check sizes
-        if not mask.shape == init_latents.shape:
-            raise ValueError("The mask and init_image should be the same size!")
+        # 5. set timesteps
+        self.scheduler.set_timesteps(num_inference_steps, device=device)
+        timesteps = self.get_timesteps(num_inference_steps, strength, device)
+        latent_timestep = timesteps[:1].repeat(batch_size * num_images_per_prompt)
 
-        # get the original timestep using init_timestep
-        offset = self.scheduler.config.get("steps_offset", 0)
-        init_timestep = int(num_inference_steps * strength) + offset
-        init_timestep = min(init_timestep, num_inference_steps)
-
-        timesteps = self.scheduler.timesteps[-init_timestep]
-        timesteps = torch.tensor([timesteps] * batch_size * num_images_per_prompt, device=self.device)
-
-        # add noise to latents using the timesteps
-        noise = torch.randn(init_latents.shape, generator=generator, device=self.device, dtype=latents_dtype)
-        init_latents = self.scheduler.add_noise(init_latents, noise, timesteps)
-
-        # prepare extra kwargs for the scheduler step, since not all schedulers have the same signature
-        # eta (η) is only used with the DDIMScheduler, it will be ignored for other schedulers.
-        # eta corresponds to η in DDIM paper: https://arxiv.org/abs/2010.02502
-        # and should be between [0, 1]
-        accepts_eta = "eta" in set(inspect.signature(self.scheduler.step).parameters.keys())
-        extra_step_kwargs = {}
-        if accepts_eta:
-            extra_step_kwargs["eta"] = eta
-
-        # check if the scheduler accepts generator
-        accepts_generator = "generator" in set(inspect.signature(self.scheduler.step).parameters.keys())
-        if accepts_generator:
-            extra_step_kwargs["generator"] = generator
-
-        latents = init_latents
+        # 6. Prepare latent variables
+        # encode the init image into latents and scale the latents
+        latents, init_latents_orig, noise = self.prepare_latents(
+            init_image, latent_timestep, batch_size, num_images_per_prompt, text_embeddings.dtype, device, generator
+        )
 
-        t_start = max(num_inference_steps - init_timestep + offset, 0)
+        # 7. Prepare mask latent
+        mask = mask_image.to(device=self.device, dtype=latents.dtype)
+        mask = torch.cat([mask] * batch_size * num_images_per_prompt)
 
-        # Some schedulers like PNDM have timesteps as arrays
-        # It's more optimized to move all timesteps to correct device beforehand
-        timesteps = self.scheduler.timesteps[t_start:].to(self.device)
+        # 8. 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)
 
-        for i, t in tqdm(enumerate(timesteps)):
+        # 9. Denoising loop
+        for i, t in enumerate(self.progress_bar(timesteps)):
             # expand the latents if we are doing classifier free guidance
             latent_model_input = torch.cat([latents] * 2) if do_classifier_free_guidance else latents
             latent_model_input = self.scheduler.scale_model_input(latent_model_input, t)
@@ -468,22 +557,13 @@ class StableDiffusionInpaintPipelineLegacy(DiffusionPipeline):
             if callback is not None and i % callback_steps == 0:
                 callback(i, t, latents)
 
-        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()
+        # 10. Post-processing
+        image = self.decode_latents(latents)
 
-        if self.safety_checker is not None:
-            safety_checker_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_checker_input.pixel_values.to(text_embeddings.dtype)
-            )
-        else:
-            has_nsfw_concept = None
+        # 11. Run safety checker
+        image, has_nsfw_concept = self.run_safety_checker(image, device, text_embeddings.dtype)
 
+        # 12. Convert to PIL
         if output_type == "pil":
             image = self.numpy_to_pil(image)
 

tests/pipelines/stable_diffusion/test_stable_diffusion_img2img.py

@@ -22,6 +22,7 @@ import torch
 
 from diffusers import (
     AutoencoderKL,
+    DDIMScheduler,
     LMSDiscreteScheduler,
     PNDMScheduler,
     StableDiffusionImg2ImgPipeline,
@@ -479,7 +480,7 @@ class StableDiffusionImg2ImgPipelineIntegrationTests(unittest.TestCase):
         )
         init_image = init_image.resize((768, 512))
         expected_image = load_numpy(
-            "https://huggingface.co/datasets/lewington/expected-images/resolve/main/fantasy_landscape.npy"
+            "https://huggingface.co/datasets/hf-internal-testing/diffusers-images/resolve/main/img2img/fantasy_landscape.npy"
         )
 
         model_id = "CompVis/stable-diffusion-v1-4"
@@ -506,7 +507,7 @@ class StableDiffusionImg2ImgPipelineIntegrationTests(unittest.TestCase):
 
         assert image.shape == (512, 768, 3)
         # img2img is flaky across GPUs even in fp32, so using MAE here
-        assert np.abs(expected_image - image).mean() < 1e-3
+        assert np.abs(expected_image - image).max() < 1e-3
 
     def test_stable_diffusion_img2img_pipeline_k_lms(self):
         init_image = load_image(
@@ -515,7 +516,7 @@ class StableDiffusionImg2ImgPipelineIntegrationTests(unittest.TestCase):
         )
         init_image = init_image.resize((768, 512))
         expected_image = load_numpy(
-            "https://huggingface.co/datasets/lewington/expected-images/resolve/main/fantasy_landscape_k_lms.npy"
+            "https://huggingface.co/datasets/hf-internal-testing/diffusers-images/resolve/main/img2img/fantasy_landscape_k_lms.npy"
         )
 
         model_id = "CompVis/stable-diffusion-v1-4"
@@ -543,8 +544,44 @@ class StableDiffusionImg2ImgPipelineIntegrationTests(unittest.TestCase):
         image = output.images[0]
 
         assert image.shape == (512, 768, 3)
-        # img2img is flaky across GPUs even in fp32, so using MAE here
-        assert np.abs(expected_image - image).mean() < 1e-3
+        assert np.abs(expected_image - image).max() < 1e-3
+
+    def test_stable_diffusion_img2img_pipeline_ddim(self):
+        init_image = load_image(
+            "https://huggingface.co/datasets/hf-internal-testing/diffusers-images/resolve/main"
+            "/img2img/sketch-mountains-input.jpg"
+        )
+        init_image = init_image.resize((768, 512))
+        expected_image = load_numpy(
+            "https://huggingface.co/datasets/hf-internal-testing/diffusers-images/resolve/main/img2img/fantasy_landscape_ddim.npy"
+        )
+
+        model_id = "CompVis/stable-diffusion-v1-4"
+        ddim = DDIMScheduler.from_config(model_id, subfolder="scheduler")
+        pipe = StableDiffusionImg2ImgPipeline.from_pretrained(
+            model_id,
+            scheduler=ddim,
+            safety_checker=None,
+        )
+        pipe.to(torch_device)
+        pipe.set_progress_bar_config(disable=None)
+        pipe.enable_attention_slicing()
+
+        prompt = "A fantasy landscape, trending on artstation"
+
+        generator = torch.Generator(device=torch_device).manual_seed(0)
+        output = pipe(
+            prompt=prompt,
+            init_image=init_image,
+            strength=0.75,
+            guidance_scale=7.5,
+            generator=generator,
+            output_type="np",
+        )
+        image = output.images[0]
+
+        assert image.shape == (512, 768, 3)
+        assert np.abs(expected_image - image).max() < 1e-3
 
     def test_stable_diffusion_img2img_intermediate_state(self):
         number_of_steps = 0

tests/pipelines/stable_diffusion/test_stable_diffusion_inpaint_legacy.py

@@ -387,7 +387,6 @@ class StableDiffusionInpaintLegacyPipelineIntegrationTests(unittest.TestCase):
         assert np.abs(expected_image - image).max() < 1e-3
 
     def test_stable_diffusion_inpaint_legacy_pipeline_k_lms(self):
-        # TODO(Anton, Patrick) - I think we can remove this test soon
         init_image = load_image(
             "https://huggingface.co/datasets/hf-internal-testing/diffusers-images/resolve/main"
             "/in_paint/overture-creations-5sI6fQgYIuo.png"