stable_diffusion_controlnet_inpaint_img2img.py

# Copyright 2023 The HuggingFace Team. All rights reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
#     http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.


import inspect
import os
import warnings
from typing import Any, Callable, Dict, List, Optional, Tuple, Union

import numpy as np
import PIL.Image
import torch
import torch.nn.functional as F
from transformers import CLIPImageProcessor, CLIPTextModel, CLIPTokenizer

from ...image_processor import VaeImageProcessor
from ...loaders import TextualInversionLoaderMixin
from ...models import AutoencoderKL, ControlNetModel, UNet2DConditionModel
from ...schedulers import KarrasDiffusionSchedulers

from ...utils import (
    PIL_INTERPOLATION,
    is_accelerate_available,
    is_accelerate_version,
    is_compiled_module,
    logging,
    randn_tensor,
    replace_example_docstring,
)
from ..pipeline_utils import DiffusionPipeline
from ..stable_diffusion import StableDiffusionPipelineOutput
from ..stable_diffusion.safety_checker import StableDiffusionSafetyChecker
from .multicontrolnet import MultiControlNetModel


logger = logging.get_logger(__name__)  # pylint: disable=invalid-name


EXAMPLE_DOC_STRING = """
    Examples:
        ```py
        >>> # !pip install opencv-python transformers accelerate
        >>> from diffusers import StableDiffusionControlNetImg2ImgPipeline, ControlNetModel, UniPCMultistepScheduler
        >>> from diffusers.utils import load_image
        >>> import numpy as np
        >>> import torch

        >>> import cv2
        >>> from PIL import Image

        >>> # download an image
        >>> image = load_image(
        ...     "https://hf.co/datasets/huggingface/documentation-images/resolve/main/diffusers/input_image_vermeer.png"
        ... )
        >>> np_image = np.array(image)

        >>> # get canny image
        >>> np_image = cv2.Canny(np_image, 100, 200)
        >>> np_image = np_image[:, :, None]
        >>> np_image = np.concatenate([np_image, np_image, np_image], axis=2)
        >>> canny_image = Image.fromarray(np_image)

        >>> # load control net and stable diffusion v1-5
        >>> controlnet = ControlNetModel.from_pretrained("lllyasviel/sd-controlnet-canny", torch_dtype=torch.float16)
        >>> pipe = StableDiffusionControlNetImg2ImgPipeline.from_pretrained(
        ...     "runwayml/stable-diffusion-v1-5", controlnet=controlnet, torch_dtype=torch.float16
        ... )

        >>> # speed up diffusion process with faster scheduler and memory optimization
        >>> pipe.scheduler = UniPCMultistepScheduler.from_config(pipe.scheduler.config)
        >>> pipe.enable_model_cpu_offload()

        >>> # generate image
        >>> generator = torch.manual_seed(0)
        >>> image = pipe(
        ...     "futuristic-looking woman",
        ...     num_inference_steps=20,
        ...     generator=generator,
        ...     image=image,
        ...     control_image=canny_image,
        ... ).images[0]
        ```
"""


def prepare_mask_and_masked_image(image, mask, height, width, return_image=False):
    """
    Prepares a pair (image, mask) to be consumed by the Stable Diffusion pipeline. This means that those inputs will be
    converted to ``torch.Tensor`` with shapes ``batch x channels x height x width`` where ``channels`` is ``3`` for the
    ``image`` and ``1`` for the ``mask``.

    The ``image`` will be converted to ``torch.float32`` and normalized to be in ``[-1, 1]``. The ``mask`` will be
    binarized (``mask > 0.5``) and cast to ``torch.float32`` too.

    Args:
        image (Union[np.array, PIL.Image, torch.Tensor]): The image to inpaint.
            It can be a ``PIL.Image``, or a ``height x width x 3`` ``np.array`` or a ``channels x height x width``
            ``torch.Tensor`` or a ``batch x channels x height x width`` ``torch.Tensor``.
        mask (_type_): The mask to apply to the image, i.e. regions to inpaint.
            It can be a ``PIL.Image``, or a ``height x width`` ``np.array`` or a ``1 x height x width``
            ``torch.Tensor`` or a ``batch x 1 x height x width`` ``torch.Tensor``.


    Raises:
        ValueError: ``torch.Tensor`` images should be in the ``[-1, 1]`` range. ValueError: ``torch.Tensor`` mask
        should be in the ``[0, 1]`` range. ValueError: ``mask`` and ``image`` should have the same spatial dimensions.
        TypeError: ``mask`` is a ``torch.Tensor`` but ``image`` is not
            (ot the other way around).

    Returns:
        tuple[torch.Tensor]: The pair (mask, masked_image) as ``torch.Tensor`` with 4
            dimensions: ``batch x channels x height x width``.
    """

    if image is None:
        raise ValueError("`image` input cannot be undefined.")

    if mask is None:
        raise ValueError("`mask_image` input cannot be undefined.")

    if isinstance(image, torch.Tensor):
        if not isinstance(mask, torch.Tensor):
            raise TypeError(f"`image` is a torch.Tensor but `mask` (type: {type(mask)} is not")

        # Batch single image
        if image.ndim == 3:
            assert image.shape[0] == 3, "Image outside a batch should be of shape (3, H, W)"
            image = image.unsqueeze(0)

        # Batch and add channel dim for single mask
        if mask.ndim == 2:
            mask = mask.unsqueeze(0).unsqueeze(0)

        # Batch single mask or add channel dim
        if mask.ndim == 3:
            # Single batched mask, no channel dim or single mask not batched but channel dim
            if mask.shape[0] == 1:
                mask = mask.unsqueeze(0)

            # Batched masks no channel dim
            else:
                mask = mask.unsqueeze(1)

        assert image.ndim == 4 and mask.ndim == 4, "Image and Mask must have 4 dimensions"
        assert image.shape[-2:] == mask.shape[-2:], "Image and Mask must have the same spatial dimensions"
        assert image.shape[0] == mask.shape[0], "Image and Mask must have the same batch size"

        # Check image is in [-1, 1]
        if image.min() < -1 or image.max() > 1:
            raise ValueError("Image should be in [-1, 1] range")

        # Check mask is in [0, 1]
        if mask.min() < 0 or mask.max() > 1:
            raise ValueError("Mask should be in [0, 1] range")

        # Binarize mask
        mask[mask < 0.5] = 0
        mask[mask >= 0.5] = 1

        # Image as float32
        image = image.to(dtype=torch.float32)
    elif isinstance(mask, torch.Tensor):
        raise TypeError(f"`mask` is a torch.Tensor but `image` (type: {type(image)} is not")
    else:
        # preprocess image
        if isinstance(image, (PIL.Image.Image, np.ndarray)):
            image = [image]
        if isinstance(image, list) and isinstance(image[0], PIL.Image.Image):
            # resize all images w.r.t passed height an width
            image = [i.resize((width, height), resample=PIL.Image.LANCZOS) for i in image]
            image = [np.array(i.convert("RGB"))[None, :] for i in image]
            image = np.concatenate(image, axis=0)
        elif isinstance(image, list) and isinstance(image[0], np.ndarray):
            image = np.concatenate([i[None, :] for i in image], axis=0)

        image = image.transpose(0, 3, 1, 2)
        image = torch.from_numpy(image).to(dtype=torch.float32) / 127.5 - 1.0

        # preprocess mask
        if isinstance(mask, (PIL.Image.Image, np.ndarray)):
            mask = [mask]

        if isinstance(mask, list) and isinstance(mask[0], PIL.Image.Image):
            mask = [i.resize((width, height), resample=PIL.Image.LANCZOS) for i in mask]
            mask = np.concatenate([np.array(m.convert("L"))[None, None, :] for m in mask], axis=0)
            mask = mask.astype(np.float32) / 255.0
        elif isinstance(mask, list) and isinstance(mask[0], np.ndarray):
            mask = np.concatenate([m[None, None, :] for m in mask], axis=0)

        mask[mask < 0.5] = 0
        mask[mask >= 0.5] = 1
        mask = torch.from_numpy(mask)

    masked_image = image * (mask < 0.5)

    # n.b. ensure backwards compatibility as old function does not return image
    if return_image:
        return mask, masked_image, image

    return mask, masked_image


class StableDiffusionControlNetImg2ImgPipeline(DiffusionPipeline, TextualInversionLoaderMixin):
    r"""
    Pipeline for text-to-image generation using Stable Diffusion with ControlNet guidance.

    This model inherits from [`DiffusionPipeline`]. Check the superclass documentation for the generic methods the
    library implements for all the pipelines (such as downloading or saving, running on a particular device, etc.)

    In addition the pipeline inherits the following loading methods:
        - *Textual-Inversion*: [`loaders.TextualInversionLoaderMixin.load_textual_inversion`]

    Args:
        vae ([`AutoencoderKL`]):
            Variational Auto-Encoder (VAE) Model to encode and decode images to and from latent representations.
        text_encoder ([`CLIPTextModel`]):
            Frozen text-encoder. Stable Diffusion uses the text portion of
            [CLIP](https://huggingface.co/docs/transformers/model_doc/clip#transformers.CLIPTextModel), specifically
            the [clip-vit-large-patch14](https://huggingface.co/openai/clip-vit-large-patch14) variant.
        tokenizer (`CLIPTokenizer`):
            Tokenizer of class
            [CLIPTokenizer](https://huggingface.co/docs/transformers/v4.21.0/en/model_doc/clip#transformers.CLIPTokenizer).
        unet ([`UNet2DConditionModel`]): Conditional U-Net architecture to denoise the encoded image latents.
        controlnet ([`ControlNetModel`] or `List[ControlNetModel]`):
            Provides additional conditioning to the unet during the denoising process. If you set multiple ControlNets
            as a list, the outputs from each ControlNet are added together to create one combined additional
            conditioning.
        scheduler ([`SchedulerMixin`]):
            A scheduler to be used in combination with `unet` to denoise the encoded image latents. Can be one of
            [`DDIMScheduler`], [`LMSDiscreteScheduler`], or [`PNDMScheduler`].
        safety_checker ([`StableDiffusionSafetyChecker`]):
            Classification module that estimates whether generated images could be considered offensive or harmful.
            Please, refer to the [model card](https://huggingface.co/runwayml/stable-diffusion-v1-5) for details.
        feature_extractor ([`CLIPImageProcessor`]):
            Model that extracts features from generated images to be used as inputs for the `safety_checker`.
    """
    _optional_components = ["safety_checker", "feature_extractor"]

    def __init__(
        self,
        vae: AutoencoderKL,
        text_encoder: CLIPTextModel,
        tokenizer: CLIPTokenizer,
        unet: UNet2DConditionModel,
        controlnet: Union[ControlNetModel, List[ControlNetModel], Tuple[ControlNetModel], MultiControlNetModel],
        scheduler: KarrasDiffusionSchedulers,
        safety_checker: StableDiffusionSafetyChecker,
        feature_extractor: CLIPImageProcessor,
        requires_safety_checker: bool = True,
    ):
        super().__init__()

        if safety_checker is None and requires_safety_checker:
            logger.warning(
                f"You have disabled the safety checker for {self.__class__} by passing `safety_checker=None`. Ensure"
                " that you abide to the conditions of the Stable Diffusion license and do not expose unfiltered"
                " results in services or applications open to the public. Both the diffusers team and Hugging Face"
                " strongly recommend to keep the safety filter enabled in all public facing circumstances, disabling"
                " it only for use-cases that involve analyzing network behavior or auditing its results. For more"
                " information, please have a look at https://github.com/huggingface/diffusers/pull/254 ."
            )

        if safety_checker is not None and feature_extractor is None:
            raise ValueError(
                "Make sure to define a feature extractor when loading {self.__class__} if you want to use the safety"
                " checker. If you do not want to use the safety checker, you can pass `'safety_checker=None'` instead."
            )

        if isinstance(controlnet, (list, tuple)):
            controlnet = MultiControlNetModel(controlnet)

        self.register_modules(
            vae=vae,
            text_encoder=text_encoder,
            tokenizer=tokenizer,
            unet=unet,
            controlnet=controlnet,
            scheduler=scheduler,
            safety_checker=safety_checker,
            feature_extractor=feature_extractor,
        )
        self.vae_scale_factor = 2 ** (len(self.vae.config.block_out_channels) - 1)
        self.image_processor = VaeImageProcessor(vae_scale_factor=self.vae_scale_factor)
        self.register_to_config(requires_safety_checker=requires_safety_checker)

    # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.enable_vae_slicing
    def enable_vae_slicing(self):
        r"""
        Enable sliced VAE decoding.

        When this option is enabled, the VAE will split the input tensor in slices to compute decoding in several
        steps. This is useful to save some memory and allow larger batch sizes.
        """
        self.vae.enable_slicing()

    # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.disable_vae_slicing
    def disable_vae_slicing(self):
        r"""
        Disable sliced VAE decoding. If `enable_vae_slicing` was previously invoked, this method will go back to
        computing decoding in one step.
        """
        self.vae.disable_slicing()

    # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.enable_vae_tiling
    def enable_vae_tiling(self):
        r"""
        Enable tiled VAE decoding.

        When this option is enabled, the VAE will split the input tensor into tiles to compute decoding and encoding in
        several steps. This is useful to save a large amount of memory and to allow the processing of larger images.
        """
        self.vae.enable_tiling()

    # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.disable_vae_tiling
    def disable_vae_tiling(self):
        r"""
        Disable tiled VAE decoding. If `enable_vae_tiling` was previously invoked, this method will go back to
        computing decoding in one step.
        """
        self.vae.disable_tiling()

    def enable_sequential_cpu_offload(self, gpu_id=0):
        r"""
        Offloads all models to CPU using accelerate, significantly reducing memory usage. When called, unet,
        text_encoder, vae, controlnet, 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.
        Note that offloading happens on a submodule basis. Memory savings are higher than with
        `enable_model_cpu_offload`, but performance is lower.
        """
        if is_accelerate_available():
            from accelerate import cpu_offload
        else:
            raise ImportError("Please install accelerate via `pip install accelerate`")

        device = torch.device(f"cuda:{gpu_id}")

        for cpu_offloaded_model in [self.unet, self.text_encoder, self.vae, self.controlnet]:
            cpu_offload(cpu_offloaded_model, device)

        if self.safety_checker is not None:
            cpu_offload(self.safety_checker, execution_device=device, offload_buffers=True)

    def enable_model_cpu_offload(self, gpu_id=0):
        r"""
        Offloads all models to CPU using accelerate, reducing memory usage with a low impact on performance. Compared
        to `enable_sequential_cpu_offload`, this method moves one whole model at a time to the GPU when its `forward`
        method is called, and the model remains in GPU until the next model runs. Memory savings are lower than with
        `enable_sequential_cpu_offload`, but performance is much better due to the iterative execution of the `unet`.
        """
        if is_accelerate_available() and is_accelerate_version(">=", "0.17.0.dev0"):
            from accelerate import cpu_offload_with_hook
        else:
            raise ImportError("`enable_model_cpu_offload` requires `accelerate v0.17.0` or higher.")

        device = torch.device(f"cuda:{gpu_id}")

        hook = None
        for cpu_offloaded_model in [self.text_encoder, self.unet, self.vae]:
            _, hook = cpu_offload_with_hook(cpu_offloaded_model, device, prev_module_hook=hook)

        if self.safety_checker is not None:
            # the safety checker can offload the vae again
            _, hook = cpu_offload_with_hook(self.safety_checker, device, prev_module_hook=hook)

        # control net hook has be manually offloaded as it alternates with unet
        cpu_offload_with_hook(self.controlnet, device)

        # We'll offload the last model manually.
        self.final_offload_hook = hook

    @property
    # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline._execution_device
    def _execution_device(self):
        r"""
        Returns the device on which the pipeline's models will be executed. After calling
        `pipeline.enable_sequential_cpu_offload()` the execution device can only be inferred from Accelerate's module
        hooks.
        """
        if not hasattr(self.unet, "_hf_hook"):
            return self.device
        for module in self.unet.modules():
            if (
                hasattr(module, "_hf_hook")
                and hasattr(module._hf_hook, "execution_device")
                and module._hf_hook.execution_device is not None
            ):
                return torch.device(module._hf_hook.execution_device)
        return self.device

    # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline._encode_prompt
    def _encode_prompt(
        self,
        prompt,
        device,
        num_images_per_prompt,
        do_classifier_free_guidance,
        negative_prompt=None,
        prompt_embeds: Optional[torch.FloatTensor] = None,
        negative_prompt_embeds: Optional[torch.FloatTensor] = None,
    ):
        r"""
        Encodes the prompt into text encoder hidden states.

        Args:
             prompt (`str` or `List[str]`, *optional*):
                prompt to be encoded
            device: (`torch.device`):
                torch device
            num_images_per_prompt (`int`):
                number of images that should be generated per prompt
            do_classifier_free_guidance (`bool`):
                whether to use classifier free guidance or not
            negative_prompt (`str` or `List[str]`, *optional*):
                The prompt or prompts not to guide the image generation. If not defined, one has to pass
                `negative_prompt_embeds` instead. Ignored when not using guidance (i.e., ignored if `guidance_scale` is
                less than `1`).
            prompt_embeds (`torch.FloatTensor`, *optional*):
                Pre-generated text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt weighting. If not
                provided, text embeddings will be generated from `prompt` input argument.
            negative_prompt_embeds (`torch.FloatTensor`, *optional*):
                Pre-generated negative text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt
                weighting. If not provided, negative_prompt_embeds will be generated from `negative_prompt` input
                argument.
        """
        if prompt is not None and isinstance(prompt, str):
            batch_size = 1
        elif prompt is not None and isinstance(prompt, list):
            batch_size = len(prompt)
        else:
            batch_size = prompt_embeds.shape[0]

        if prompt_embeds is None:
            # textual inversion: procecss multi-vector tokens if necessary
            if isinstance(self, TextualInversionLoaderMixin):
                prompt = self.maybe_convert_prompt(prompt, self.tokenizer)

            text_inputs = self.tokenizer(
                prompt,
                padding="max_length",
                max_length=self.tokenizer.model_max_length,
                truncation=True,
                return_tensors="pt",
            )
            text_input_ids = text_inputs.input_ids
            untruncated_ids = self.tokenizer(prompt, padding="longest", return_tensors="pt").input_ids

            if untruncated_ids.shape[-1] >= text_input_ids.shape[-1] and not torch.equal(
                text_input_ids, untruncated_ids
            ):
                removed_text = self.tokenizer.batch_decode(
                    untruncated_ids[:, self.tokenizer.model_max_length - 1 : -1]
                )
                logger.warning(
                    "The following part of your input was truncated because CLIP can only handle sequences up to"
                    f" {self.tokenizer.model_max_length} tokens: {removed_text}"
                )

            if hasattr(self.text_encoder.config, "use_attention_mask") and self.text_encoder.config.use_attention_mask:
                attention_mask = text_inputs.attention_mask.to(device)
            else:
                attention_mask = None

            prompt_embeds = self.text_encoder(
                text_input_ids.to(device),
                attention_mask=attention_mask,
            )
            prompt_embeds = prompt_embeds[0]

        prompt_embeds = prompt_embeds.to(dtype=self.text_encoder.dtype, device=device)

        bs_embed, seq_len, _ = prompt_embeds.shape
        # duplicate text embeddings for each generation per prompt, using mps friendly method
        prompt_embeds = prompt_embeds.repeat(1, num_images_per_prompt, 1)
        prompt_embeds = prompt_embeds.view(bs_embed * num_images_per_prompt, seq_len, -1)

        # get unconditional embeddings for classifier free guidance
        if do_classifier_free_guidance and negative_prompt_embeds is None:
            uncond_tokens: List[str]
            if negative_prompt is None:
                uncond_tokens = [""] * batch_size
            elif prompt is not None and type(prompt) is not type(negative_prompt):
                raise TypeError(
                    f"`negative_prompt` should be the same type to `prompt`, but got {type(negative_prompt)} !="
                    f" {type(prompt)}."
                )
            elif isinstance(negative_prompt, str):
                uncond_tokens = [negative_prompt]
            elif batch_size != len(negative_prompt):
                raise ValueError(
                    f"`negative_prompt`: {negative_prompt} has batch size {len(negative_prompt)}, but `prompt`:"
                    f" {prompt} has batch size {batch_size}. Please make sure that passed `negative_prompt` matches"
                    " the batch size of `prompt`."
                )
            else:
                uncond_tokens = negative_prompt

            # textual inversion: procecss multi-vector tokens if necessary
            if isinstance(self, TextualInversionLoaderMixin):
                uncond_tokens = self.maybe_convert_prompt(uncond_tokens, self.tokenizer)

            max_length = prompt_embeds.shape[1]
            uncond_input = self.tokenizer(
                uncond_tokens,
                padding="max_length",
                max_length=max_length,
                truncation=True,
                return_tensors="pt",
            )

            if hasattr(self.text_encoder.config, "use_attention_mask") and self.text_encoder.config.use_attention_mask:
                attention_mask = uncond_input.attention_mask.to(device)
            else:
                attention_mask = None

            negative_prompt_embeds = self.text_encoder(
                uncond_input.input_ids.to(device),
                attention_mask=attention_mask,
            )
            negative_prompt_embeds = negative_prompt_embeds[0]

        if do_classifier_free_guidance:
            # duplicate unconditional embeddings for each generation per prompt, using mps friendly method
            seq_len = negative_prompt_embeds.shape[1]

            negative_prompt_embeds = negative_prompt_embeds.to(dtype=self.text_encoder.dtype, device=device)

            negative_prompt_embeds = negative_prompt_embeds.repeat(1, num_images_per_prompt, 1)
            negative_prompt_embeds = negative_prompt_embeds.view(batch_size * num_images_per_prompt, seq_len, -1)

            # 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
            prompt_embeds = torch.cat([negative_prompt_embeds, prompt_embeds])

        return prompt_embeds

    # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.run_safety_checker
    def run_safety_checker(self, image, device, dtype):
        if self.safety_checker is None:
            has_nsfw_concept = None
        else:
            if torch.is_tensor(image):
                feature_extractor_input = self.image_processor.postprocess(image, output_type="pil")
            else:
                feature_extractor_input = self.image_processor.numpy_to_pil(image)
            safety_checker_input = self.feature_extractor(feature_extractor_input, return_tensors="pt").to(device)
            image, has_nsfw_concept = self.safety_checker(
                images=image, clip_input=safety_checker_input.pixel_values.to(dtype)
            )
        return image, has_nsfw_concept

    # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.decode_latents
    def decode_latents(self, latents):
        warnings.warn(
            "The decode_latents method is deprecated and will be removed in a future version. Please"
            " use VaeImageProcessor instead",
            FutureWarning,
        )
        latents = 1 / self.vae.config.scaling_factor * latents
        image = self.vae.decode(latents, return_dict=False)[0]
        image = (image / 2 + 0.5).clamp(0, 1)
        # we always cast to float32 as this does not cause significant overhead and is compatible with bfloat16
        image = image.cpu().permute(0, 2, 3, 1).float().numpy()
        return image

    # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.prepare_extra_step_kwargs
    def prepare_extra_step_kwargs(self, generator, eta):
        # prepare extra kwargs for the scheduler step, since not all schedulers have the same signature
        # eta (η) is only used with the DDIMScheduler, it will be ignored for other schedulers.
        # eta corresponds to η in DDIM paper: https://arxiv.org/abs/2010.02502
        # and should be between [0, 1]

        accepts_eta = "eta" in set(inspect.signature(self.scheduler.step).parameters.keys())
        extra_step_kwargs = {}
        if accepts_eta:
            extra_step_kwargs["eta"] = eta

        # check if the scheduler accepts generator
        accepts_generator = "generator" in set(inspect.signature(self.scheduler.step).parameters.keys())
        if accepts_generator:
            extra_step_kwargs["generator"] = generator
        return extra_step_kwargs
    
        # 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
        init_timestep = min(int(num_inference_steps * strength), num_inference_steps)

        t_start = max(num_inference_steps - init_timestep, 0)
        timesteps = self.scheduler.timesteps[t_start * self.scheduler.order :]

        return timesteps, num_inference_steps - t_start

    def check_inputs(
        self,
        prompt,
        image,
        height,
        width,
        callback_steps,
        negative_prompt=None,
        prompt_embeds=None,
        negative_prompt_embeds=None,
        controlnet_conditioning_scale=1.0,
    ):
        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)}."
            )

        if prompt is not None and prompt_embeds is not None:
            raise ValueError(
                f"Cannot forward both `prompt`: {prompt} and `prompt_embeds`: {prompt_embeds}. Please make sure to"
                " only forward one of the two."
            )
        elif prompt is None and prompt_embeds is None:
            raise ValueError(
                "Provide either `prompt` or `prompt_embeds`. Cannot leave both `prompt` and `prompt_embeds` undefined."
            )
        elif prompt is not None and (not isinstance(prompt, str) and not isinstance(prompt, list)):
            raise ValueError(f"`prompt` has to be of type `str` or `list` but is {type(prompt)}")

        if negative_prompt is not None and negative_prompt_embeds is not None:
            raise ValueError(
                f"Cannot forward both `negative_prompt`: {negative_prompt} and `negative_prompt_embeds`:"
                f" {negative_prompt_embeds}. Please make sure to only forward one of the two."
            )

        if prompt_embeds is not None and negative_prompt_embeds is not None:
            if prompt_embeds.shape != negative_prompt_embeds.shape:
                raise ValueError(
                    "`prompt_embeds` and `negative_prompt_embeds` must have the same shape when passed directly, but"
                    f" got: `prompt_embeds` {prompt_embeds.shape} != `negative_prompt_embeds`"
                    f" {negative_prompt_embeds.shape}."
                )

        # `prompt` needs more sophisticated handling when there are multiple
        # conditionings.
        if isinstance(self.controlnet, MultiControlNetModel):
            if isinstance(prompt, list):
                logger.warning(
                    f"You have {len(self.controlnet.nets)} ControlNets and you have passed {len(prompt)}"
                    " prompts. The conditionings will be fixed across the prompts."
                )

        # Check `image`
        is_compiled = hasattr(F, "scaled_dot_product_attention") and isinstance(
            self.controlnet, torch._dynamo.eval_frame.OptimizedModule
        )
        if (
            isinstance(self.controlnet, ControlNetModel)
            or is_compiled
            and isinstance(self.controlnet._orig_mod, ControlNetModel)
        ):
            self.check_image(image, prompt, prompt_embeds)
        elif (
            isinstance(self.controlnet, MultiControlNetModel)
            or is_compiled
            and isinstance(self.controlnet._orig_mod, MultiControlNetModel)
        ):
            if not isinstance(image, list):
                raise TypeError("For multiple controlnets: `image` must be type `list`")

            # When `image` is a nested list:
            # (e.g. [[canny_image_1, pose_image_1], [canny_image_2, pose_image_2]])
            elif any(isinstance(i, list) for i in image):
                raise ValueError("A single batch of multiple conditionings are supported at the moment.")
            elif len(image) != len(self.controlnet.nets):
                raise ValueError(
                    "For multiple controlnets: `image` must have the same length as the number of controlnets."
                )

            for image_ in image:
                self.check_image(image_, prompt, prompt_embeds)
        else:
            assert False

        # Check `controlnet_conditioning_scale`
        if (
            isinstance(self.controlnet, ControlNetModel)
            or is_compiled
            and isinstance(self.controlnet._orig_mod, ControlNetModel)
        ):
            if not isinstance(controlnet_conditioning_scale, float):
                raise TypeError("For single controlnet: `controlnet_conditioning_scale` must be type `float`.")
        elif (
            isinstance(self.controlnet, MultiControlNetModel)
            or is_compiled
            and isinstance(self.controlnet._orig_mod, MultiControlNetModel)
        ):
            if isinstance(controlnet_conditioning_scale, list):
                if any(isinstance(i, list) for i in controlnet_conditioning_scale):
                    raise ValueError("A single batch of multiple conditionings are supported at the moment.")
            elif isinstance(controlnet_conditioning_scale, list) and len(controlnet_conditioning_scale) != len(
                self.controlnet.nets
            ):
                raise ValueError(
                    "For multiple controlnets: When `controlnet_conditioning_scale` is specified as `list`, it must have"
                    " the same length as the number of controlnets"
                )
        else:
            assert False

    def check_image(self, image, prompt, prompt_embeds):
        image_is_pil = isinstance(image, PIL.Image.Image)
        image_is_tensor = isinstance(image, torch.Tensor)
        image_is_pil_list = isinstance(image, list) and isinstance(image[0], PIL.Image.Image)
        image_is_tensor_list = isinstance(image, list) and isinstance(image[0], torch.Tensor)

        if not image_is_pil and not image_is_tensor and not image_is_pil_list and not image_is_tensor_list:
            raise TypeError(
                "image must be passed and be one of PIL image, torch tensor, list of PIL images, or list of torch tensors"
            )

        if image_is_pil:
            image_batch_size = 1
        elif image_is_tensor:
            image_batch_size = image.shape[0]
        elif image_is_pil_list:
            image_batch_size = len(image)
        elif image_is_tensor_list:
            image_batch_size = len(image)

        if prompt is not None and isinstance(prompt, str):
            prompt_batch_size = 1
        elif prompt is not None and isinstance(prompt, list):
            prompt_batch_size = len(prompt)
        elif prompt_embeds is not None:
            prompt_batch_size = prompt_embeds.shape[0]

        if image_batch_size != 1 and image_batch_size != prompt_batch_size:
            raise ValueError(
                f"If image batch size is not 1, image batch size must be same as prompt batch size. image batch size: {image_batch_size}, prompt batch size: {prompt_batch_size}"
            )

    # Copied from diffusers.pipelines.controlnet.pipeline_controlnet.StableDiffusionControlNetPipeline.prepare_image
    def prepare_control_image(
        self,
        image,
        width,
        height,
        batch_size,
        num_images_per_prompt,
        device,
        dtype,
        do_classifier_free_guidance=False,
        guess_mode=False,
    ):
        if not isinstance(image, torch.Tensor):
            if isinstance(image, PIL.Image.Image):
                image = [image]

            if isinstance(image[0], PIL.Image.Image):
                images = []

                for image_ in image:
                    image_ = image_.convert("RGB")
                    image_ = image_.resize((width, height), resample=PIL_INTERPOLATION["lanczos"])
                    image_ = np.array(image_)
                    image_ = image_[None, :]
                    images.append(image_)

                image = images

                image = np.concatenate(image, axis=0)
                image = np.array(image).astype(np.float32) / 255.0
                image = image.transpose(0, 3, 1, 2)
                image = torch.from_numpy(image)
            elif isinstance(image[0], torch.Tensor):
                image = torch.cat(image, dim=0)

        image_batch_size = image.shape[0]

        if image_batch_size == 1:
            repeat_by = batch_size
        else:
            # image batch size is the same as prompt batch size
            repeat_by = num_images_per_prompt

        image = image.repeat_interleave(repeat_by, dim=0)

        image = image.to(device=device, dtype=dtype)

        if do_classifier_free_guidance and not guess_mode:
            image = torch.cat([image] * 2)

        return image

    # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion_img2img.StableDiffusionImg2ImgPipeline.get_timesteps
    def get_timesteps(self, num_inference_steps, strength, device):
        # get the original timestep using init_timestep
        init_timestep = min(int(num_inference_steps * strength), num_inference_steps)

        t_start = max(num_inference_steps - init_timestep, 0)
        timesteps = self.scheduler.timesteps[t_start * self.scheduler.order :]

        return timesteps, num_inference_steps - t_start

    # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion_img2img.StableDiffusionImg2ImgPipeline.prepare_latents
    def prepare_latents(
        self,
        batch_size,
        num_channels_latents,
        height,
        width,
        dtype,
        device,
        generator,
        latents=None,
        image=None,
        timestep=None,
        is_strength_max=True,
        return_noise=False,
        return_image_latents=False,
    ):
        shape = (batch_size, num_channels_latents, height // self.vae_scale_factor, width // self.vae_scale_factor)
        if isinstance(generator, list) and len(generator) != batch_size:
            raise ValueError(
                f"You have passed a list of generators of length {len(generator)}, but requested an effective batch"
                f" size of {batch_size}. Make sure the batch size matches the length of the generators."
            )

        if (image is None or timestep is None) and not is_strength_max:
            raise ValueError(
                "Since strength < 1. initial latents are to be initialised as a combination of Image + Noise."
                "However, either the image or the noise timestep has not been provided."
            )

        if return_image_latents or (latents is None and not is_strength_max):
            image = image.to(device=device, dtype=dtype)
            image_latents = self._encode_vae_image(image=image, generator=generator)

        if latents is None:
            noise = randn_tensor(shape, generator=generator, device=device, dtype=dtype)
            # if strength is 1. then initialise the latents to noise, else initial to image + noise
            latents = noise if is_strength_max else self.scheduler.add_noise(image_latents, noise, timestep)
            # if pure noise then scale the initial latents by the  Scheduler's init sigma
            latents = latents * self.scheduler.init_noise_sigma if is_strength_max else latents
        else:
            noise = latents.to(device)
            latents = noise * self.scheduler.init_noise_sigma

        outputs = (latents,)

        if return_noise:
            outputs += (noise,)

        if return_image_latents:
            outputs += (image_latents,)

        return outputs

    def _default_height_width(self, height, width, image):
        # NOTE: It is possible that a list of images have different
        # dimensions for each image, so just checking the first image
        # is not _exactly_ correct, but it is simple.
        while isinstance(image, list):
            image = image[0]

        if height is None:
            if isinstance(image, PIL.Image.Image):
                height = image.height
            elif isinstance(image, torch.Tensor):
                height = image.shape[2]

            height = (height // 8) * 8  # round down to nearest multiple of 8

        if width is None:
            if isinstance(image, PIL.Image.Image):
                width = image.width
            elif isinstance(image, torch.Tensor):
                width = image.shape[3]

            width = (width // 8) * 8  # round down to nearest multiple of 8

        return height, width
    
     # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion_inpaint.StableDiffusionInpaintPipeline.prepare_mask_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 // self.vae_scale_factor, width // self.vae_scale_factor)
        )
        mask = mask.to(device=device, dtype=dtype)

        masked_image = masked_image.to(device=device, dtype=dtype)
        masked_image_latents = self._encode_vae_image(masked_image, generator=generator)

        # duplicate mask and masked_image_latents for each generation per prompt, using mps friendly method
        if mask.shape[0] < batch_size:
            if not batch_size % mask.shape[0] == 0:
                raise ValueError(
                    "The passed mask and the required batch size don't match. Masks are supposed to be duplicated to"
                    f" a total batch size of {batch_size}, but {mask.shape[0]} masks were passed. Make sure the number"
                    " of masks that you pass is divisible by the total requested batch size."
                )
            mask = mask.repeat(batch_size // mask.shape[0], 1, 1, 1)
        if masked_image_latents.shape[0] < batch_size:
            if not batch_size % masked_image_latents.shape[0] == 0:
                raise ValueError(
                    "The passed images and the required batch size don't match. Images are supposed to be duplicated"
                    f" to a total batch size of {batch_size}, but {masked_image_latents.shape[0]} images were passed."
                    " Make sure the number of images that you pass is divisible by the total requested batch size."
                )
            masked_image_latents = masked_image_latents.repeat(batch_size // masked_image_latents.shape[0], 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
    
    def _encode_vae_image(self, image: torch.Tensor, generator: torch.Generator):
        if isinstance(generator, list):
            image_latents = [
                self.vae.encode(image[i : i + 1]).latent_dist.sample(generator=generator[i])
                for i in range(image.shape[0])
            ]
            image_latents = torch.cat(image_latents, dim=0)
        else:
            image_latents = self.vae.encode(image).latent_dist.sample(generator=generator)

        image_latents = self.vae.config.scaling_factor * image_latents

        return image_latents

    # override DiffusionPipeline
    def save_pretrained(
        self,
        save_directory: Union[str, os.PathLike],
        safe_serialization: bool = False,
        variant: Optional[str] = None,
    ):
        if isinstance(self.controlnet, ControlNetModel):
            super().save_pretrained(save_directory, safe_serialization, variant)
        else:
            raise NotImplementedError("Currently, the `save_pretrained()` is not implemented for Multi-ControlNet.")

    @torch.no_grad()
    @replace_example_docstring(EXAMPLE_DOC_STRING)
    def __call__(
        self,
        prompt: Union[str, List[str]] = None,
        image: Union[torch.FloatTensor, PIL.Image.Image, List[torch.FloatTensor], List[PIL.Image.Image]] = None,
        mask_image: Union[torch.Tensor, PIL.Image.Image] = None,
        control_image: Union[
            torch.FloatTensor, PIL.Image.Image, List[torch.FloatTensor], List[PIL.Image.Image]
        ] = None,
        height: Optional[int] = None,
        width: Optional[int] = None,
        strength: float = 0.8,
        num_inference_steps: int = 50,
        guidance_scale: float = 7.5,
        negative_prompt: Optional[Union[str, List[str]]] = None,
        num_images_per_prompt: Optional[int] = 1,
        eta: float = 0.0,
        generator: Optional[Union[torch.Generator, List[torch.Generator]]] = None,
        latents: Optional[torch.FloatTensor] = None,
        prompt_embeds: Optional[torch.FloatTensor] = None,
        negative_prompt_embeds: Optional[torch.FloatTensor] = None,
        output_type: Optional[str] = "pil",
        return_dict: bool = True,
        callback: Optional[Callable[[int, int, torch.FloatTensor], None]] = None,
        callback_steps: int = 1,
        cross_attention_kwargs: Optional[Dict[str, Any]] = None,
        controlnet_conditioning_scale: Union[float, List[float]] = 0.8,
        guess_mode: bool = False,
    ):
        r"""
        Function invoked when calling the pipeline for generation.

        Args:
            prompt (`str` or `List[str]`, *optional*):
                The prompt or prompts to guide the image generation. If not defined, one has to pass `prompt_embeds`.
                instead.
            image (`torch.FloatTensor`, `PIL.Image.Image`, `List[torch.FloatTensor]`, `List[PIL.Image.Image]`,
                    `List[List[torch.FloatTensor]]`, or `List[List[PIL.Image.Image]]`):
                The ControlNet input condition. ControlNet uses this input condition to generate guidance to Unet. If
                the type is specified as `Torch.FloatTensor`, it is passed to ControlNet as is. `PIL.Image.Image` can
                also be accepted as an image. The dimensions of the output image defaults to `image`'s dimensions. If
                height and/or width are passed, `image` is resized according to them. If multiple ControlNets are
                specified in init, images must be passed as a list such that each element of the list can be correctly
                batched for input to a single controlnet.
            height (`int`, *optional*, defaults to self.unet.config.sample_size * self.vae_scale_factor):
                The height in pixels of the generated image.
            width (`int`, *optional*, defaults to self.unet.config.sample_size * self.vae_scale_factor):
                The width in pixels of the generated image.
            num_inference_steps (`int`, *optional*, defaults to 50):
                The number of denoising steps. More denoising steps usually lead to a higher quality image at the
                expense of slower inference.
            guidance_scale (`float`, *optional*, defaults to 7.5):
                Guidance scale as defined in [Classifier-Free Diffusion Guidance](https://arxiv.org/abs/2207.12598).
                `guidance_scale` is defined as `w` of equation 2. of [Imagen
                Paper](https://arxiv.org/pdf/2205.11487.pdf). Guidance scale is enabled by setting `guidance_scale >
                1`. Higher guidance scale encourages to generate images that are closely linked to the text `prompt`,
                usually at the expense of lower image quality.
            negative_prompt (`str` or `List[str]`, *optional*):
                The prompt or prompts not to guide the image generation. If not defined, one has to pass
                `negative_prompt_embeds` instead. Ignored when not using guidance (i.e., ignored if `guidance_scale` is
                less than `1`).
            num_images_per_prompt (`int`, *optional*, defaults to 1):
                The number of images to generate per prompt.
            eta (`float`, *optional*, defaults to 0.0):
                Corresponds to parameter eta (η) in the DDIM paper: https://arxiv.org/abs/2010.02502. Only applies to
                [`schedulers.DDIMScheduler`], will be ignored for others.
            generator (`torch.Generator` or `List[torch.Generator]`, *optional*):
                One or a list of [torch generator(s)](https://pytorch.org/docs/stable/generated/torch.Generator.html)
                to make generation deterministic.
            latents (`torch.FloatTensor`, *optional*):
                Pre-generated noisy latents, sampled from a Gaussian distribution, to be used as inputs for image
                generation. Can be used to tweak the same generation with different prompts. If not provided, a latents
                tensor will ge generated by sampling using the supplied random `generator`.
            prompt_embeds (`torch.FloatTensor`, *optional*):
                Pre-generated text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt weighting. If not
                provided, text embeddings will be generated from `prompt` input argument.
            negative_prompt_embeds (`torch.FloatTensor`, *optional*):
                Pre-generated negative text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt
                weighting. If not provided, negative_prompt_embeds will be generated from `negative_prompt` input
                argument.
            output_type (`str`, *optional*, defaults to `"pil"`):
                The output format of the generate image. Choose between
                [PIL](https://pillow.readthedocs.io/en/stable/): `PIL.Image.Image` or `np.array`.
            return_dict (`bool`, *optional*, defaults to `True`):
                Whether or not to return a [`~pipelines.stable_diffusion.StableDiffusionPipelineOutput`] instead of a
                plain tuple.
            callback (`Callable`, *optional*):
                A function that will be called every `callback_steps` steps during inference. The function will be
                called with the following arguments: `callback(step: int, timestep: int, latents: torch.FloatTensor)`.
            callback_steps (`int`, *optional*, defaults to 1):
                The frequency at which the `callback` function will be called. If not specified, the callback will be
                called at every step.
            cross_attention_kwargs (`dict`, *optional*):
                A kwargs dictionary that if specified is passed along to the `AttentionProcessor` as defined under
                `self.processor` in
                [diffusers.cross_attention](https://github.com/huggingface/diffusers/blob/main/src/diffusers/models/cross_attention.py).
            controlnet_conditioning_scale (`float` or `List[float]`, *optional*, defaults to 1.0):
                The outputs of the controlnet are multiplied by `controlnet_conditioning_scale` before they are added
                to the residual in the original unet. If multiple ControlNets are specified in init, you can set the
                corresponding scale as a list. Note that by default, we use a smaller conditioning scale for inpainting
                than for [`~StableDiffusionControlNetPipeline.__call__`].
            guess_mode (`bool`, *optional*, defaults to `False`):
                In this mode, the ControlNet encoder will try best to recognize the content of the input image even if
                you remove all prompts. The `guidance_scale` between 3.0 and 5.0 is recommended.

        Examples:

        Returns:
            [`~pipelines.stable_diffusion.StableDiffusionPipelineOutput`] or `tuple`:
            [`~pipelines.stable_diffusion.StableDiffusionPipelineOutput`] if `return_dict` is True, otherwise a `tuple.
            When returning a tuple, the first element is a list with the generated images, and the second element is a
            list of `bool`s denoting whether the corresponding generated image likely represents "not-safe-for-work"
            (nsfw) content, according to the `safety_checker`.
        """
        # 0. Default height and width to unet
        height, width = self._default_height_width(height, width, image)

        # 1. Check inputs. Raise error if not correct
        self.check_inputs(
            prompt,
            control_image,
            height,
            width,
            callback_steps,
            negative_prompt,
            prompt_embeds,
            negative_prompt_embeds,
            controlnet_conditioning_scale,
        )

        # 2. Define call parameters
        if prompt is not None and isinstance(prompt, str):
            batch_size = 1
        elif prompt is not None and isinstance(prompt, list):
            batch_size = len(prompt)
        else:
            batch_size = prompt_embeds.shape[0]

        device = self._execution_device
        # 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

        controlnet = self.controlnet._orig_mod if is_compiled_module(self.controlnet) else self.controlnet

        if isinstance(controlnet, MultiControlNetModel) and isinstance(controlnet_conditioning_scale, float):
            controlnet_conditioning_scale = [controlnet_conditioning_scale] * len(controlnet.nets)

        global_pool_conditions = (
            controlnet.config.global_pool_conditions
            if isinstance(controlnet, ControlNetModel)
            else controlnet.nets[0].config.global_pool_conditions
        )
        guess_mode = guess_mode or global_pool_conditions

        # 3. Encode input prompt
        prompt_embeds = self._encode_prompt(
            prompt,
            device,
            num_images_per_prompt,
            do_classifier_free_guidance,
            negative_prompt,
            prompt_embeds=prompt_embeds,
            negative_prompt_embeds=negative_prompt_embeds,
        )
        # 4. Prepare image, and controlnet_conditioning_image
        #image = prepare_image(image)

        # 5. Prepare image
        if isinstance(controlnet, ControlNetModel):
            control_image = self.prepare_control_image(
                image=control_image,
                width=width,
                height=height,
                batch_size=batch_size * num_images_per_prompt,
                num_images_per_prompt=num_images_per_prompt,
                device=device,
                dtype=controlnet.dtype,
                do_classifier_free_guidance=do_classifier_free_guidance,
                guess_mode=guess_mode,
            )
        elif isinstance(controlnet, MultiControlNetModel):
            control_images = []

            for control_image_ in control_image:
                control_image_ = self.prepare_control_image(
                    image=control_image_,
                    width=width,
                    height=height,
                    batch_size=batch_size * num_images_per_prompt,
                    num_images_per_prompt=num_images_per_prompt,
                    device=device,
                    dtype=controlnet.dtype,
                    do_classifier_free_guidance=do_classifier_free_guidance,
                    guess_mode=guess_mode,
                )

                control_images.append(control_image_)

            control_image = control_images
        else:
            assert False

        # 4. Preprocess mask and image - resizes image and mask w.r.t height and width
        mask, masked_image, init_image = prepare_mask_and_masked_image(
            image, mask_image, height, width, return_image=True
        )
        # 5. Prepare timesteps
        self.scheduler.set_timesteps(num_inference_steps, device=device)
        timesteps, num_inference_steps = self.get_timesteps(num_inference_steps, strength, device)
        latent_timestep = timesteps[:1].repeat(batch_size * num_images_per_prompt)

        is_strength_max = strength == 1.0

        # 6. Prepare latent variables
        num_channels_latents = self.vae.config.latent_channels
        num_channels_unet = self.unet.config.in_channels
        return_image_latents = num_channels_unet == 4
        latents_outputs = self.prepare_latents(
            batch_size * num_images_per_prompt,
            num_channels_latents,
            height,
            width,
            prompt_embeds.dtype,
            device,
            generator,
            latents,
            image=init_image,
            timestep=latent_timestep,
            is_strength_max=is_strength_max,
            return_noise=True,
            return_image_latents=return_image_latents,
        )

        if return_image_latents:
            latents, noise, image_latents = latents_outputs
        else:
            latents, noise = latents_outputs

        # 7. Prepare mask latent variables
        mask, masked_image_latents = self.prepare_mask_latents(
            mask,
            masked_image,
            batch_size * num_images_per_prompt,
            height,
            width,
            prompt_embeds.dtype,
            device,
            generator,
            do_classifier_free_guidance,
        )

        # 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
        num_warmup_steps = len(timesteps) - num_inference_steps * self.scheduler.order
        with self.progress_bar(total=num_inference_steps) as progress_bar:
            for i, t in enumerate(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)

                # controlnet(s) inference
                if guess_mode and do_classifier_free_guidance:
                    # Infer ControlNet only for the conditional batch.
                    control_model_input = latents
                    control_model_input = self.scheduler.scale_model_input(control_model_input, t)
                    controlnet_prompt_embeds = prompt_embeds.chunk(2)[1]
                else:
                    control_model_input = latent_model_input
                    controlnet_prompt_embeds = prompt_embeds

                down_block_res_samples, mid_block_res_sample = self.controlnet(
                    control_model_input,
                    t,
                    encoder_hidden_states=controlnet_prompt_embeds,
                    controlnet_cond=control_image,
                    conditioning_scale=controlnet_conditioning_scale,
                    guess_mode=guess_mode,
                    return_dict=False,
                )

                if guess_mode and do_classifier_free_guidance:
                    # Infered ControlNet only for the conditional batch.
                    # To apply the output of ControlNet to both the unconditional and conditional batches,
                    # add 0 to the unconditional batch to keep it unchanged.
                    down_block_res_samples = [torch.cat([torch.zeros_like(d), d]) for d in down_block_res_samples]
                    mid_block_res_sample = torch.cat([torch.zeros_like(mid_block_res_sample), mid_block_res_sample])

                
                # predict the noise residual
                if num_channels_unet == 9:
                    latent_model_input = torch.cat([latent_model_input, mask, masked_image_latents], dim=1)

                # predict the noise residual
                noise_pred = self.unet(
                    latent_model_input,
                    t,
                    encoder_hidden_states=prompt_embeds,
                    cross_attention_kwargs=cross_attention_kwargs,
                    down_block_additional_residuals=down_block_res_samples,
                    mid_block_additional_residual=mid_block_res_sample,
                    return_dict=False,
                )[0]

                # 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
                latents = self.scheduler.step(noise_pred, t, latents, **extra_step_kwargs, return_dict=False)[0]

                
                if num_channels_unet == 4:
                    init_latents_proper = image_latents[:1]
                    init_mask = mask[:1]

                    if i < len(timesteps) - 1:
                        init_latents_proper = self.scheduler.add_noise(init_latents_proper, noise, torch.tensor([t]))

                    latents = (1 - init_mask) * init_latents_proper + init_mask * latents

                # call the callback, if provided
                if i == len(timesteps) - 1 or ((i + 1) > num_warmup_steps and (i + 1) % self.scheduler.order == 0):
                    progress_bar.update()
                    if callback is not None and i % callback_steps == 0:
                        callback(i, t, latents)

        # If we do sequential model offloading, let's offload unet and controlnet
        # manually for max memory savings
        if hasattr(self, "final_offload_hook") and self.final_offload_hook is not None:
            self.unet.to("cpu")
            self.controlnet.to("cpu")
            torch.cuda.empty_cache()

        if not output_type == "latent":
            image = self.vae.decode(latents / self.vae.config.scaling_factor, return_dict=False)[0]
            image, has_nsfw_concept = self.run_safety_checker(image, device, prompt_embeds.dtype)
        else:
            image = latents
            has_nsfw_concept = None

        if has_nsfw_concept is None:
            do_denormalize = [True] * image.shape[0]
        else:
            do_denormalize = [not has_nsfw for has_nsfw in has_nsfw_concept]

        image = self.image_processor.postprocess(image, output_type=output_type, do_denormalize=do_denormalize)

        # Offload last model to CPU
        if hasattr(self, "final_offload_hook") and self.final_offload_hook is not None:
            self.final_offload_hook.offload()

        if not return_dict:
            return (image, has_nsfw_concept)

        return StableDiffusionPipelineOutput(images=image, nsfw_content_detected=has_nsfw_concept)