pipeline_controlnet_xs_sd_xl_instantid.py

import inspect
import copy, os 
from safetensors.torch import load_file
from typing import Any, Callable, Dict, List, Optional, Tuple, Union
import collections
import numpy as np
import PIL.Image
import torch
import torch.nn.functional as F
from transformers import (
    CLIPImageProcessor,
    CLIPTextModel,
    CLIPTextModelWithProjection,
    CLIPTokenizer,
)
import gc
from diffusers.utils.import_utils import is_xformers_available
from diffusers.utils.import_utils import is_invisible_watermark_available

from diffusers.image_processor import PipelineImageInput, VaeImageProcessor
from diffusers.loaders import FromSingleFileMixin, StableDiffusionXLLoraLoaderMixin, TextualInversionLoaderMixin
from diffusers.models import AutoencoderKL
from diffusers.models.attention_processor import (
    AttnProcessor2_0,
    LoRAAttnProcessor2_0,
    LoRAXFormersAttnProcessor,
    XFormersAttnProcessor,
)
from diffusers.models.lora import adjust_lora_scale_text_encoder
from diffusers.schedulers import KarrasDiffusionSchedulers

from diffusers.utils import (
    USE_PEFT_BACKEND,
    is_torch_version,
    logging,
    replace_example_docstring,
    scale_lora_layers,
    unscale_lora_layers,
    delete_adapter_layers,
    set_adapter_layers,
    set_weights_and_activate_adapters,
)

from diffusers.utils.torch_utils import is_compiled_module, is_torch_version, randn_tensor
from diffusers.pipelines.pipeline_utils import DiffusionPipeline
from diffusers.pipelines.stable_diffusion_xl.pipeline_output import StableDiffusionXLPipelineOutput


from utils.callbacks import MultiPipelineCallbacks, PipelineCallback

# lora
from diffusers.models.unets.unet_2d_condition import UNet2DConditionModel
from controlnet_xs import ControlNetXSAdapter, UNetControlNetXSModel
from diffusers.loaders.lora_conversion_utils import  _maybe_map_sgm_blocks_to_diffusers, _convert_non_diffusers_lora_to_diffusers
from utils.tools import get_module_kohya_state_dict_xs


#ipa
from ip_adapter.resampler import Resampler
from ip_adapter.utils import is_torch2_available
if is_torch2_available():
    from ip_adapter.attention_processor import IPAttnProcessor2_0 as IPAttnProcessor, AttnProcessor2_0 as AttnProcessor
else:
    from ip_adapter.attention_processor import IPAttnProcessor, AttnProcessor
from ip_adapter.attention_processor import region_control


if is_invisible_watermark_available():
    from diffusers.pipelines.stable_diffusion_xl.watermark import StableDiffusionXLWatermarker


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


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

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

        >>> prompt = "aerial view, a futuristic research complex in a bright foggy jungle, hard lighting"
        >>> negative_prompt = "low quality, bad quality, sketches"

        >>> # download an image
        >>> image = load_image(
        ...     "https://hf.co/datasets/hf-internal-testing/diffusers-images/resolve/main/sd_controlnet/hf-logo.png"
        ... )

        >>> # initialize the models and pipeline
        >>> controlnet_conditioning_scale = 0.5
        >>> vae = AutoencoderKL.from_pretrained("madebyollin/sdxl-vae-fp16-fix", torch_dtype=torch.float16)
        >>> controlnet = ControlNetXSAdapter.from_pretrained(
        ...     "UmerHA/Testing-ConrolNetXS-SDXL-canny", torch_dtype=torch.float16
        ... )
        >>> pipe = StableDiffusionXLControlNetXSPipeline.from_pretrained(
        ...     "stabilityai/stable-diffusion-xl-base-1.0", controlnet=controlnet, torch_dtype=torch.float16
        ... )
        >>> pipe.enable_model_cpu_offload()

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

        >>> # generate image
        >>> image = pipe(
        ...     prompt, controlnet_conditioning_scale=controlnet_conditioning_scale, image=canny_image
        ... ).images[0]
        ```
"""


from transformers import CLIPTokenizer
from diffusers.pipelines.stable_diffusion_xl import StableDiffusionXLPipeline

class LongPromptWeight(object):
    """
    Copied from https://github.com/huggingface/diffusers/blob/main/examples/community/lpw_stable_diffusion_xl.py
    """

    def __init__(self) -> None:
        pass

    def parse_prompt_attention(self, text):
        """
        Parses a string with attention tokens and returns a list of pairs: text and its associated weight.
        Accepted tokens are:
        (abc) - increases attention to abc by a multiplier of 1.1
        (abc:3.12) - increases attention to abc by a multiplier of 3.12
        [abc] - decreases attention to abc by a multiplier of 1.1
        \( - literal character '('
        \[ - literal character '['
        \) - literal character ')'
        \] - literal character ']'
        \\ - literal character '\'
        anything else - just text

        >>> parse_prompt_attention('normal text')
        [['normal text', 1.0]]
        >>> parse_prompt_attention('an (important) word')
        [['an ', 1.0], ['important', 1.1], [' word', 1.0]]
        >>> parse_prompt_attention('(unbalanced')
        [['unbalanced', 1.1]]
        >>> parse_prompt_attention('\(literal\]')
        [['(literal]', 1.0]]
        >>> parse_prompt_attention('(unnecessary)(parens)')
        [['unnecessaryparens', 1.1]]
        >>> parse_prompt_attention('a (((house:1.3)) [on] a (hill:0.5), sun, (((sky))).')
        [['a ', 1.0],
        ['house', 1.5730000000000004],
        [' ', 1.1],
        ['on', 1.0],
        [' a ', 1.1],
        ['hill', 0.55],
        [', sun, ', 1.1],
        ['sky', 1.4641000000000006],
        ['.', 1.1]]
        """
        import re

        re_attention = re.compile(
            r"""
                \\\(|\\\)|\\\[|\\]|\\\\|\\|\(|\[|:([+-]?[.\d]+)\)|
                \)|]|[^\\()\[\]:]+|:
            """,
            re.X,
        )

        re_break = re.compile(r"\s*\bBREAK\b\s*", re.S)

        res = []
        round_brackets = []
        square_brackets = []

        round_bracket_multiplier = 1.1
        square_bracket_multiplier = 1 / 1.1

        def multiply_range(start_position, multiplier):
            for p in range(start_position, len(res)):
                res[p][1] *= multiplier

        for m in re_attention.finditer(text):
            text = m.group(0)
            weight = m.group(1)

            if text.startswith("\\"):
                res.append([text[1:], 1.0])
            elif text == "(":
                round_brackets.append(len(res))
            elif text == "[":
                square_brackets.append(len(res))
            elif weight is not None and len(round_brackets) > 0:
                multiply_range(round_brackets.pop(), float(weight))
            elif text == ")" and len(round_brackets) > 0:
                multiply_range(round_brackets.pop(), round_bracket_multiplier)
            elif text == "]" and len(square_brackets) > 0:
                multiply_range(square_brackets.pop(), square_bracket_multiplier)
            else:
                parts = re.split(re_break, text)
                for i, part in enumerate(parts):
                    if i > 0:
                        res.append(["BREAK", -1])
                    res.append([part, 1.0])

        for pos in round_brackets:
            multiply_range(pos, round_bracket_multiplier)

        for pos in square_brackets:
            multiply_range(pos, square_bracket_multiplier)

        if len(res) == 0:
            res = [["", 1.0]]

        # merge runs of identical weights
        i = 0
        while i + 1 < len(res):
            if res[i][1] == res[i + 1][1]:
                res[i][0] += res[i + 1][0]
                res.pop(i + 1)
            else:
                i += 1

        return res

    def get_prompts_tokens_with_weights(self, clip_tokenizer: CLIPTokenizer, prompt: str):
        """
        Get prompt token ids and weights, this function works for both prompt and negative prompt

        Args:
            pipe (CLIPTokenizer)
                A CLIPTokenizer
            prompt (str)
                A prompt string with weights

        Returns:
            text_tokens (list)
                A list contains token ids
            text_weight (list)
                A list contains the correspodent weight of token ids

        Example:
            import torch
            from transformers import CLIPTokenizer

            clip_tokenizer = CLIPTokenizer.from_pretrained(
                "stablediffusionapi/deliberate-v2"
                , subfolder = "tokenizer"
                , dtype = torch.float16
            )

            token_id_list, token_weight_list = get_prompts_tokens_with_weights(
                clip_tokenizer = clip_tokenizer
                ,prompt = "a (red:1.5) cat"*70
            )
        """
        texts_and_weights = self.parse_prompt_attention(prompt)
        text_tokens, text_weights = [], []
        for word, weight in texts_and_weights:
            # tokenize and discard the starting and the ending token
            token = clip_tokenizer(word, truncation=False).input_ids[1:-1]  # so that tokenize whatever length prompt
            # the returned token is a 1d list: [320, 1125, 539, 320]

            # merge the new tokens to the all tokens holder: text_tokens
            text_tokens = [*text_tokens, *token]

            # each token chunk will come with one weight, like ['red cat', 2.0]
            # need to expand weight for each token.
            chunk_weights = [weight] * len(token)

            # append the weight back to the weight holder: text_weights
            text_weights = [*text_weights, *chunk_weights]
        return text_tokens, text_weights

    def group_tokens_and_weights(self, token_ids: list, weights: list, pad_last_block=False):
        """
        Produce tokens and weights in groups and pad the missing tokens

        Args:
            token_ids (list)
                The token ids from tokenizer
            weights (list)
                The weights list from function get_prompts_tokens_with_weights
            pad_last_block (bool)
                Control if fill the last token list to 75 tokens with eos
        Returns:
            new_token_ids (2d list)
            new_weights (2d list)

        Example:
            token_groups,weight_groups = group_tokens_and_weights(
                token_ids = token_id_list
                , weights = token_weight_list
            )
        """
        bos, eos = 49406, 49407

        # this will be a 2d list
        new_token_ids = []
        new_weights = []
        while len(token_ids) >= 75:
            # get the first 75 tokens
            head_75_tokens = [token_ids.pop(0) for _ in range(75)]
            head_75_weights = [weights.pop(0) for _ in range(75)]

            # extract token ids and weights
            temp_77_token_ids = [bos] + head_75_tokens + [eos]
            temp_77_weights = [1.0] + head_75_weights + [1.0]

            # add 77 token and weights chunk to the holder list
            new_token_ids.append(temp_77_token_ids)
            new_weights.append(temp_77_weights)

        # padding the left
        if len(token_ids) >= 0:
            padding_len = 75 - len(token_ids) if pad_last_block else 0

            temp_77_token_ids = [bos] + token_ids + [eos] * padding_len + [eos]
            new_token_ids.append(temp_77_token_ids)

            temp_77_weights = [1.0] + weights + [1.0] * padding_len + [1.0]
            new_weights.append(temp_77_weights)

        return new_token_ids, new_weights

    def get_weighted_text_embeddings_sdxl(
            self,
            pipe: StableDiffusionXLPipeline,
            prompt: str = "",
            prompt_2: str = None,
            neg_prompt: str = "",
            neg_prompt_2: str = None,
            prompt_embeds=None,
            negative_prompt_embeds=None,
            pooled_prompt_embeds=None,
            negative_pooled_prompt_embeds=None,
            extra_emb=None,
            extra_emb_alpha=0.6,
    ):
        """
        This function can process long prompt with weights, no length limitation
        for Stable Diffusion XL

        Args:
            pipe (StableDiffusionPipeline)
            prompt (str)
            prompt_2 (str)
            neg_prompt (str)
            neg_prompt_2 (str)
        Returns:
            prompt_embeds (torch.Tensor)
            neg_prompt_embeds (torch.Tensor)
        """
        # 
        if prompt_embeds is not None and \
                negative_prompt_embeds is not None and \
                pooled_prompt_embeds is not None and \
                negative_pooled_prompt_embeds is not None:
            return prompt_embeds, negative_prompt_embeds, pooled_prompt_embeds, negative_pooled_prompt_embeds

        if prompt_2:
            prompt = f"{prompt} {prompt_2}"

        if neg_prompt_2:
            neg_prompt = f"{neg_prompt} {neg_prompt_2}"

        eos = pipe.tokenizer.eos_token_id

        # tokenizer 1
        prompt_tokens, prompt_weights = self.get_prompts_tokens_with_weights(pipe.tokenizer, prompt)
        neg_prompt_tokens, neg_prompt_weights = self.get_prompts_tokens_with_weights(pipe.tokenizer, neg_prompt)

        # tokenizer 2
        # prompt_tokens_2, prompt_weights_2 = self.get_prompts_tokens_with_weights(pipe.tokenizer_2, prompt)
        # neg_prompt_tokens_2, neg_prompt_weights_2 = self.get_prompts_tokens_with_weights(pipe.tokenizer_2, neg_prompt)
        # tokenizer 2 遇到 !! !!!! 等多感叹号和tokenizer 1的效果不一致
        prompt_tokens_2, prompt_weights_2 = self.get_prompts_tokens_with_weights(pipe.tokenizer, prompt)
        neg_prompt_tokens_2, neg_prompt_weights_2 = self.get_prompts_tokens_with_weights(pipe.tokenizer, neg_prompt)

        # padding the shorter one for prompt set 1
        prompt_token_len = len(prompt_tokens)
        neg_prompt_token_len = len(neg_prompt_tokens)

        if prompt_token_len > neg_prompt_token_len:
            # padding the neg_prompt with eos token
            neg_prompt_tokens = neg_prompt_tokens + [eos] * abs(prompt_token_len - neg_prompt_token_len)
            neg_prompt_weights = neg_prompt_weights + [1.0] * abs(prompt_token_len - neg_prompt_token_len)
        else:
            # padding the prompt
            prompt_tokens = prompt_tokens + [eos] * abs(prompt_token_len - neg_prompt_token_len)
            prompt_weights = prompt_weights + [1.0] * abs(prompt_token_len - neg_prompt_token_len)

        # padding the shorter one for token set 2
        prompt_token_len_2 = len(prompt_tokens_2)
        neg_prompt_token_len_2 = len(neg_prompt_tokens_2)

        if prompt_token_len_2 > neg_prompt_token_len_2:
            # padding the neg_prompt with eos token
            neg_prompt_tokens_2 = neg_prompt_tokens_2 + [eos] * abs(prompt_token_len_2 - neg_prompt_token_len_2)
            neg_prompt_weights_2 = neg_prompt_weights_2 + [1.0] * abs(prompt_token_len_2 - neg_prompt_token_len_2)
        else:
            # padding the prompt
            prompt_tokens_2 = prompt_tokens_2 + [eos] * abs(prompt_token_len_2 - neg_prompt_token_len_2)
            prompt_weights_2 = prompt_weights + [1.0] * abs(prompt_token_len_2 - neg_prompt_token_len_2)

        embeds = []
        neg_embeds = []

        prompt_token_groups, prompt_weight_groups = self.group_tokens_and_weights(prompt_tokens.copy(),
                                                                                  prompt_weights.copy())

        neg_prompt_token_groups, neg_prompt_weight_groups = self.group_tokens_and_weights(
            neg_prompt_tokens.copy(), neg_prompt_weights.copy()
        )

        prompt_token_groups_2, prompt_weight_groups_2 = self.group_tokens_and_weights(
            prompt_tokens_2.copy(), prompt_weights_2.copy()
        )

        neg_prompt_token_groups_2, neg_prompt_weight_groups_2 = self.group_tokens_and_weights(
            neg_prompt_tokens_2.copy(), neg_prompt_weights_2.copy()
        )

        # get prompt embeddings one by one is not working.
        for i in range(len(prompt_token_groups)):
            # get positive prompt embeddings with weights
            token_tensor = torch.tensor([prompt_token_groups[i]], dtype=torch.long, device=pipe.device)
            weight_tensor = torch.tensor(prompt_weight_groups[i], dtype=torch.float16, device=pipe.device)

            token_tensor_2 = torch.tensor([prompt_token_groups_2[i]], dtype=torch.long, device=pipe.device)

            # use first text encoder
            prompt_embeds_1 = pipe.text_encoder(token_tensor.to(pipe.device), output_hidden_states=True)
            prompt_embeds_1_hidden_states = prompt_embeds_1.hidden_states[-2]

            # use second text encoder
            prompt_embeds_2 = pipe.text_encoder_2(token_tensor_2.to(pipe.device), output_hidden_states=True)
            prompt_embeds_2_hidden_states = prompt_embeds_2.hidden_states[-2]
            pooled_prompt_embeds = prompt_embeds_2[0]

            prompt_embeds_list = [prompt_embeds_1_hidden_states, prompt_embeds_2_hidden_states]
            token_embedding = torch.concat(prompt_embeds_list, dim=-1).squeeze(0)

            for j in range(len(weight_tensor)):
                if weight_tensor[j] != 1.0:
                    token_embedding[j] = (
                            token_embedding[-1] + (token_embedding[j] - token_embedding[-1]) * weight_tensor[j]
                    )

            token_embedding = token_embedding.unsqueeze(0)
            embeds.append(token_embedding)

            # get negative prompt embeddings with weights
            neg_token_tensor = torch.tensor([neg_prompt_token_groups[i]], dtype=torch.long, device=pipe.device)
            neg_token_tensor_2 = torch.tensor([neg_prompt_token_groups_2[i]], dtype=torch.long, device=pipe.device)
            neg_weight_tensor = torch.tensor(neg_prompt_weight_groups[i], dtype=torch.float16, device=pipe.device)

            # use first text encoder
            neg_prompt_embeds_1 = pipe.text_encoder(neg_token_tensor.to(pipe.device), output_hidden_states=True)
            neg_prompt_embeds_1_hidden_states = neg_prompt_embeds_1.hidden_states[-2]

            # use second text encoder
            neg_prompt_embeds_2 = pipe.text_encoder_2(neg_token_tensor_2.to(pipe.device), output_hidden_states=True)
            neg_prompt_embeds_2_hidden_states = neg_prompt_embeds_2.hidden_states[-2]
            negative_pooled_prompt_embeds = neg_prompt_embeds_2[0]

            neg_prompt_embeds_list = [neg_prompt_embeds_1_hidden_states, neg_prompt_embeds_2_hidden_states]
            neg_token_embedding = torch.concat(neg_prompt_embeds_list, dim=-1).squeeze(0)

            for z in range(len(neg_weight_tensor)):
                if neg_weight_tensor[z] != 1.0:
                    neg_token_embedding[z] = (
                            neg_token_embedding[-1] + (neg_token_embedding[z] - neg_token_embedding[-1]) *
                            neg_weight_tensor[z]
                    )

            neg_token_embedding = neg_token_embedding.unsqueeze(0)
            neg_embeds.append(neg_token_embedding)

        prompt_embeds = torch.cat(embeds, dim=1)
        negative_prompt_embeds = torch.cat(neg_embeds, dim=1)

        if extra_emb is not None:
            extra_emb = extra_emb.to(prompt_embeds.device, dtype=prompt_embeds.dtype) * extra_emb_alpha
            prompt_embeds = torch.cat([prompt_embeds, extra_emb], 1)
            negative_prompt_embeds = torch.cat([negative_prompt_embeds, torch.zeros_like(extra_emb)], 1)
            print(f'fix prompt_embeds, extra_emb_alpha={extra_emb_alpha}')

        return prompt_embeds, negative_prompt_embeds, pooled_prompt_embeds, negative_pooled_prompt_embeds

    def get_prompt_embeds(self, *args, **kwargs):
        prompt_embeds, negative_prompt_embeds, _, _ = self.get_weighted_text_embeddings_sdxl(*args, **kwargs)
        prompt_embeds = torch.cat([negative_prompt_embeds, prompt_embeds], dim=0)
        return prompt_embeds


class StableDiffusionXLControlNetXSPipeline(
    DiffusionPipeline,
    TextualInversionLoaderMixin,
    StableDiffusionXLLoraLoaderMixin,
    FromSingleFileMixin,
):
    r"""
    Pipeline for text-to-image generation using Stable Diffusion XL with ControlNet-XS guidance.

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

    The pipeline also inherits the following loading methods:
        - [`~loaders.TextualInversionLoaderMixin.load_textual_inversion`] for loading textual inversion embeddings
        - [`loaders.StableDiffusionXLLoraLoaderMixin.load_lora_weights`] for loading LoRA weights
        - [`loaders.FromSingleFileMixin.from_single_file`] for loading `.ckpt` files

    Args:
        vae ([`AutoencoderKL`]):
            Variational Auto-Encoder (VAE) model to encode and decode images to and from latent representations.
        text_encoder ([`~transformers.CLIPTextModel`]):
            Frozen text-encoder ([clip-vit-large-patch14](https://huggingface.co/openai/clip-vit-large-patch14)).
        text_encoder_2 ([`~transformers.CLIPTextModelWithProjection`]):
            Second frozen text-encoder
            ([laion/CLIP-ViT-bigG-14-laion2B-39B-b160k](https://huggingface.co/laion/CLIP-ViT-bigG-14-laion2B-39B-b160k)).
        tokenizer ([`~transformers.CLIPTokenizer`]):
            A `CLIPTokenizer` to tokenize text.
        tokenizer_2 ([`~transformers.CLIPTokenizer`]):
            A `CLIPTokenizer` to tokenize text.
        unet ([`UNet2DConditionModel`]):
            A [`UNet2DConditionModel`] used to create a UNetControlNetXSModel to denoise the encoded image latents.
        controlnet ([`ControlNetXSAdapter`]):
            A [`ControlNetXSAdapter`] to be used in combination with `unet` to denoise the encoded image latents.
        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`].
        force_zeros_for_empty_prompt (`bool`, *optional*, defaults to `"True"`):
            Whether the negative prompt embeddings should always be set to 0. Also see the config of
            `stabilityai/stable-diffusion-xl-base-1-0`.
        add_watermarker (`bool`, *optional*):
            Whether to use the [invisible_watermark](https://github.com/ShieldMnt/invisible-watermark/) library to
            watermark output images. If not defined, it defaults to `True` if the package is installed; otherwise no
            watermarker is used.
    """

    model_cpu_offload_seq = "text_encoder->text_encoder_2->unet->vae"
    _optional_components = [
        "tokenizer",
        "tokenizer_2",
        "text_encoder",
        "text_encoder_2",
        "feature_extractor",
    ]
    _callback_tensor_inputs = [
        "latents",
        "prompt_embeds",
        "negative_prompt_embeds",
        "add_text_embeds",
        "add_time_ids",
        "negative_pooled_prompt_embeds",
        "negative_add_time_ids",
    ]

    def __init__(
        self,
        vae: AutoencoderKL,
        text_encoder: CLIPTextModel,
        text_encoder_2: CLIPTextModelWithProjection,
        tokenizer: CLIPTokenizer,
        tokenizer_2: CLIPTokenizer,
        unet: Union[UNet2DConditionModel, UNetControlNetXSModel],
        controlnet: ControlNetXSAdapter,
        scheduler: KarrasDiffusionSchedulers,
        force_zeros_for_empty_prompt: bool = True,
        add_watermarker: Optional[bool] = None,
        feature_extractor: CLIPImageProcessor = None,
    ):
        super().__init__()
        # self.org_unet_config = copy.deepcopy(unet.config)
        if isinstance(unet, UNet2DConditionModel):
            unet = UNetControlNetXSModel.from_unet(unet, controlnet)

        self.register_modules(
            vae=vae,
            text_encoder=text_encoder,
            text_encoder_2=text_encoder_2,
            tokenizer=tokenizer,
            tokenizer_2=tokenizer_2,
            unet=unet,
            controlnet=controlnet,
            scheduler=scheduler,
            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, do_convert_rgb=True)
        self.control_image_processor = VaeImageProcessor(
            vae_scale_factor=self.vae_scale_factor, do_convert_rgb=True, do_normalize=False
        )
        add_watermarker = add_watermarker if add_watermarker is not None else is_invisible_watermark_available()

        if add_watermarker:
            self.watermark = StableDiffusionXLWatermarker()
        else:
            self.watermark = None

        self.register_to_config(force_zeros_for_empty_prompt=force_zeros_for_empty_prompt)   
    
    def cuda(self, org_unet_config=None, device='cuda', dtype=torch.float16, use_xformers=False):
        self.org_unet_config = org_unet_config
        self.to(device, dtype)
        
        if hasattr(self, 'image_proj_model'):
            self.image_proj_model.to(device).to(dtype)
        
        if use_xformers:
            if is_xformers_available():
                import xformers
                from packaging import version

                xformers_version = version.parse(xformers.__version__)
                if xformers_version == version.parse("0.0.16"):
                    logger.warn(
                        "xFormers 0.0.16 cannot be used for training in some GPUs. If you observe problems during training, please update xFormers to at least 0.0.17. See https://huggingface.co/docs/diffusers/main/en/optimization/xformers for more details."
                    )
                self.enable_xformers_memory_efficient_attention()
            else:
                raise ValueError("xformers is not available. Make sure it is installed correctly")
    
    def encode_prompt(
        self,
        prompt: str,
        prompt_2: Optional[str] = None,
        device: Optional[torch.device] = None,
        num_images_per_prompt: int = 1,
        do_classifier_free_guidance: bool = True,
        negative_prompt: Optional[str] = None,
        negative_prompt_2: Optional[str] = None,
        prompt_embeds: Optional[torch.Tensor] = None,
        negative_prompt_embeds: Optional[torch.Tensor] = None,
        pooled_prompt_embeds: Optional[torch.Tensor] = None,
        negative_pooled_prompt_embeds: Optional[torch.Tensor] = None,
        lora_scale: Optional[float] = None,
        clip_skip: Optional[int] = None,
    ):
        r"""
        Encodes the prompt into text encoder hidden states.

        Args:
            prompt (`str` or `List[str]`, *optional*):
                prompt to be encoded
            prompt_2 (`str` or `List[str]`, *optional*):
                The prompt or prompts to be sent to the `tokenizer_2` and `text_encoder_2`. If not defined, `prompt` is
                used in both text-encoders
            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`).
            negative_prompt_2 (`str` or `List[str]`, *optional*):
                The prompt or prompts not to guide the image generation to be sent to `tokenizer_2` and
                `text_encoder_2`. If not defined, `negative_prompt` is used in both text-encoders
            prompt_embeds (`torch.Tensor`, *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.Tensor`, *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.
            pooled_prompt_embeds (`torch.Tensor`, *optional*):
                Pre-generated pooled text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt weighting.
                If not provided, pooled text embeddings will be generated from `prompt` input argument.
            negative_pooled_prompt_embeds (`torch.Tensor`, *optional*):
                Pre-generated negative pooled text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt
                weighting. If not provided, pooled negative_prompt_embeds will be generated from `negative_prompt`
                input argument.
            lora_scale (`float`, *optional*):
                A lora scale that will be applied to all LoRA layers of the text encoder if LoRA layers are loaded.
            clip_skip (`int`, *optional*):
                Number of layers to be skipped from CLIP while computing the prompt embeddings. A value of 1 means that
                the output of the pre-final layer will be used for computing the prompt embeddings.
        """
        device = device or self._execution_device

        # set lora scale so that monkey patched LoRA
        # function of text encoder can correctly access it
        if lora_scale is not None and isinstance(self, StableDiffusionXLLoraLoaderMixin):
            self._lora_scale = lora_scale

            # dynamically adjust the LoRA scale
            if self.text_encoder is not None:
                if not USE_PEFT_BACKEND:
                    adjust_lora_scale_text_encoder(self.text_encoder, lora_scale)
                else:
                    scale_lora_layers(self.text_encoder, lora_scale)

            if self.text_encoder_2 is not None:
                if not USE_PEFT_BACKEND:
                    adjust_lora_scale_text_encoder(self.text_encoder_2, lora_scale)
                else:
                    scale_lora_layers(self.text_encoder_2, lora_scale)

        prompt = [prompt] if isinstance(prompt, str) else prompt
 
        if prompt is not None:
            batch_size = len(prompt)
        else:
            batch_size = prompt_embeds.shape[0]

        # Define tokenizers and text encoders
        tokenizers = [self.tokenizer, self.tokenizer_2] if self.tokenizer is not None else [self.tokenizer_2]
        text_encoders = (
            [self.text_encoder, self.text_encoder_2] if self.text_encoder is not None else [self.text_encoder_2]
        )

        if prompt_embeds is None:
            prompt_2 = prompt_2 or prompt
            prompt_2 = [prompt_2] if isinstance(prompt_2, str) else prompt_2

            # textual inversion: process multi-vector tokens if necessary
            prompt_embeds_list = []
            prompts = [prompt, prompt_2]
            for prompt, tokenizer, text_encoder in zip(prompts, tokenizers, text_encoders):
                if isinstance(self, TextualInversionLoaderMixin):
                    prompt = self.maybe_convert_prompt(prompt, tokenizer)

                text_inputs = tokenizer(
                    prompt,
                    padding="max_length",
                    max_length=tokenizer.model_max_length,
                    truncation=True,
                    return_tensors="pt",
                )

                text_input_ids = text_inputs.input_ids
                untruncated_ids = 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 = tokenizer.batch_decode(untruncated_ids[:, 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" {tokenizer.model_max_length} tokens: {removed_text}"
                    )

                prompt_embeds = text_encoder(text_input_ids.to(device), output_hidden_states=True)

                # We are only ALWAYS interested in the pooled output of the final text encoder
                pooled_prompt_embeds = prompt_embeds[0]
                if clip_skip is None:
                    prompt_embeds = prompt_embeds.hidden_states[-2]
                else:
                    # "2" because SDXL always indexes from the penultimate layer.
                    prompt_embeds = prompt_embeds.hidden_states[-(clip_skip + 2)]

                prompt_embeds_list.append(prompt_embeds)

            prompt_embeds = torch.concat(prompt_embeds_list, dim=-1)

        # get unconditional embeddings for classifier free guidance
        zero_out_negative_prompt = negative_prompt is None and self.config.force_zeros_for_empty_prompt
        if do_classifier_free_guidance and negative_prompt_embeds is None and zero_out_negative_prompt:
            negative_prompt_embeds = torch.zeros_like(prompt_embeds)
            negative_pooled_prompt_embeds = torch.zeros_like(pooled_prompt_embeds)
        elif do_classifier_free_guidance and negative_prompt_embeds is None:
            negative_prompt = negative_prompt or ""
            negative_prompt_2 = negative_prompt_2 or negative_prompt

            # normalize str to list
            negative_prompt = batch_size * [negative_prompt] if isinstance(negative_prompt, str) else negative_prompt
            negative_prompt_2 = (
                batch_size * [negative_prompt_2] if isinstance(negative_prompt_2, str) else negative_prompt_2
            )

            uncond_tokens: List[str]
            if 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 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, negative_prompt_2]

            negative_prompt_embeds_list = []
            for negative_prompt, tokenizer, text_encoder in zip(uncond_tokens, tokenizers, text_encoders):
                if isinstance(self, TextualInversionLoaderMixin):
                    negative_prompt = self.maybe_convert_prompt(negative_prompt, tokenizer)

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

                negative_prompt_embeds = text_encoder(
                    uncond_input.input_ids.to(device),
                    output_hidden_states=True,
                )
                # We are only ALWAYS interested in the pooled output of the final text encoder
                negative_pooled_prompt_embeds = negative_prompt_embeds[0]
                negative_prompt_embeds = negative_prompt_embeds.hidden_states[-2]

                negative_prompt_embeds_list.append(negative_prompt_embeds)

            negative_prompt_embeds = torch.concat(negative_prompt_embeds_list, dim=-1)

        if self.text_encoder_2 is not None:
            prompt_embeds = prompt_embeds.to(dtype=self.text_encoder_2.dtype, device=device)
        else:
            prompt_embeds = prompt_embeds.to(dtype=self.unet.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)

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

            if self.text_encoder_2 is not None:
                negative_prompt_embeds = negative_prompt_embeds.to(dtype=self.text_encoder_2.dtype, device=device)
            else:
                negative_prompt_embeds = negative_prompt_embeds.to(dtype=self.unet.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)

        pooled_prompt_embeds = pooled_prompt_embeds.repeat(1, num_images_per_prompt).view(
            bs_embed * num_images_per_prompt, -1
        )
        if do_classifier_free_guidance:
            negative_pooled_prompt_embeds = negative_pooled_prompt_embeds.repeat(1, num_images_per_prompt).view(
                bs_embed * num_images_per_prompt, -1
            )

        if self.text_encoder is not None:
            if isinstance(self, StableDiffusionXLLoraLoaderMixin) and USE_PEFT_BACKEND:
                # Retrieve the original scale by scaling back the LoRA layers
                unscale_lora_layers(self.text_encoder, lora_scale)

        if self.text_encoder_2 is not None:
            if isinstance(self, StableDiffusionXLLoraLoaderMixin) and USE_PEFT_BACKEND:
                # Retrieve the original scale by scaling back the LoRA layers
                unscale_lora_layers(self.text_encoder_2, lora_scale)

        return prompt_embeds, negative_prompt_embeds, pooled_prompt_embeds, negative_pooled_prompt_embeds

    # 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,
        prompt_2,
        image,
        negative_prompt=None,
        negative_prompt_2=None,
        prompt_embeds=None,
        negative_prompt_embeds=None,
        pooled_prompt_embeds=None,
        negative_pooled_prompt_embeds=None,
        controlnet_conditioning_scale=1.0,
        control_guidance_start=0.0,
        control_guidance_end=1.0,
        callback_on_step_end_tensor_inputs=None,
    ):
        if callback_on_step_end_tensor_inputs is not None and not all(
            k in self._callback_tensor_inputs for k in callback_on_step_end_tensor_inputs
        ):
            raise ValueError(
                f"`callback_on_step_end_tensor_inputs` has to be in {self._callback_tensor_inputs}, but found {[k for k in callback_on_step_end_tensor_inputs if k not in self._callback_tensor_inputs]}"
            )

        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_2 is not None and prompt_embeds is not None:
            raise ValueError(
                f"Cannot forward both `prompt_2`: {prompt_2} 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)}")
        elif prompt_2 is not None and (not isinstance(prompt_2, str) and not isinstance(prompt_2, list)):
            raise ValueError(f"`prompt_2` has to be of type `str` or `list` but is {type(prompt_2)}")

        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."
            )
        elif negative_prompt_2 is not None and negative_prompt_embeds is not None:
            raise ValueError(
                f"Cannot forward both `negative_prompt_2`: {negative_prompt_2} 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}."
                )

        if prompt_embeds is not None and pooled_prompt_embeds is None:
            raise ValueError(
                "If `prompt_embeds` are provided, `pooled_prompt_embeds` also have to be passed. Make sure to generate `pooled_prompt_embeds` from the same text encoder that was used to generate `prompt_embeds`."
            )

        if negative_prompt_embeds is not None and negative_pooled_prompt_embeds is None:
            raise ValueError(
                "If `negative_prompt_embeds` are provided, `negative_pooled_prompt_embeds` also have to be passed. Make sure to generate `negative_pooled_prompt_embeds` from the same text encoder that was used to generate `negative_prompt_embeds`."
            )

        # Check `image` and ``controlnet_conditioning_scale``
        is_compiled = hasattr(F, "scaled_dot_product_attention") and isinstance(
            self.unet, torch._dynamo.eval_frame.OptimizedModule
        )
        if (
            isinstance(self.unet, UNetControlNetXSModel)
            or is_compiled
            and isinstance(self.unet._orig_mod, UNetControlNetXSModel)
        ):
            self.check_image(image, prompt, prompt_embeds)
            if not isinstance(controlnet_conditioning_scale, float):
                raise TypeError("For single controlnet: `controlnet_conditioning_scale` must be type `float`.")
        else:
            assert False

        start, end = control_guidance_start, control_guidance_end
        if start >= end:
            raise ValueError(
                f"control guidance start: {start} cannot be larger or equal to control guidance end: {end}."
            )
        if start < 0.0:
            raise ValueError(f"control guidance start: {start} can't be smaller than 0.")
        if end > 1.0:
            raise ValueError(f"control guidance end: {end} can't be larger than 1.0.")

    # Copied from diffusers.pipelines.controlnet.pipeline_controlnet.StableDiffusionControlNetPipeline.check_image
    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_np = isinstance(image, np.ndarray)
        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)
        image_is_np_list = isinstance(image, list) and isinstance(image[0], np.ndarray)

        if (
            not image_is_pil
            and not image_is_tensor
            and not image_is_np
            and not image_is_pil_list
            and not image_is_tensor_list
            and not image_is_np_list
        ):
            raise TypeError(
                f"image must be passed and be one of PIL image, numpy array, torch tensor, list of PIL images, list of numpy arrays or list of torch tensors, but is {type(image)}"
            )

        if image_is_pil:
            image_batch_size = 1
        else:
            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}"
            )

    def prepare_image(
        self,
        image,
        width,
        height,
        batch_size,
        num_images_per_prompt,
        device,
        dtype,
        do_classifier_free_guidance=False,
    ):
        image = self.control_image_processor.preprocess(image, height=height, width=width).to(dtype=torch.float32)
        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:
            image = torch.cat([image] * 2)

        return image

    # 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,
            int(height) // self.vae_scale_factor,
            int(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 latents is None:
            latents = randn_tensor(shape, generator=generator, device=device, dtype=dtype)
        else:
            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 _get_add_time_ids(
        self, original_size, crops_coords_top_left, target_size, dtype, text_encoder_projection_dim=None
    ):
        add_time_ids = list(original_size + crops_coords_top_left + target_size)

        passed_add_embed_dim = (
            self.unet.config.addition_time_embed_dim * len(add_time_ids) + text_encoder_projection_dim
        )
        expected_add_embed_dim = self.unet.base_add_embedding.linear_1.in_features

        if expected_add_embed_dim != passed_add_embed_dim:
            raise ValueError(
                f"Model expects an added time embedding vector of length {expected_add_embed_dim}, but a vector of {passed_add_embed_dim} was created. The model has an incorrect config. Please check `unet.config.time_embedding_type` and `text_encoder_2.config.projection_dim`."
            )

        add_time_ids = torch.tensor([add_time_ids], dtype=dtype)
        return add_time_ids

    # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion_upscale.StableDiffusionUpscalePipeline.upcast_vae
    def upcast_vae(self):
        dtype = self.vae.dtype
        self.vae.to(dtype=torch.float32)
        use_torch_2_0_or_xformers = isinstance(
            self.vae.decoder.mid_block.attentions[0].processor,
            (
                AttnProcessor2_0,
                XFormersAttnProcessor,
                LoRAXFormersAttnProcessor,
                LoRAAttnProcessor2_0,
            ),
        )
        # if xformers or torch_2_0 is used attention block does not need
        # to be in float32 which can save lots of memory
        if use_torch_2_0_or_xformers:
            self.vae.post_quant_conv.to(dtype)
            self.vae.decoder.conv_in.to(dtype)
            self.vae.decoder.mid_block.to(dtype)

    @property
    # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.guidance_scale
    def guidance_scale(self):
        return self._guidance_scale

    @property
    # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.clip_skip
    def clip_skip(self):
        return self._clip_skip

    @property
    # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.do_classifier_free_guidance
    def do_classifier_free_guidance(self):
        return self._guidance_scale > 1 and self.unet.config.time_cond_proj_dim is None

    @property
    # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.cross_attention_kwargs
    def cross_attention_kwargs(self):
        return self._cross_attention_kwargs

    @property
    # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.num_timesteps
    def num_timesteps(self):
        return self._num_timesteps
    
    def load_ip_adapter(self, image_proj_model, cross_attn_path=None, image_emb_dim=512, num_tokens=16, device='cuda', dtype=torch.float16):     
        self.set_image_proj_model(image_proj_model, image_emb_dim, num_tokens, device=device, dtype=dtype)
        if cross_attn_path != None:
            self.set_cross_attn(cross_attn_path, num_tokens)
    
    def set_image_proj_model(self, model_ckpt, image_emb_dim=512, num_tokens=16, device='cuda', dtype=torch.float16):
        
        image_proj_model = Resampler(
            dim=1280,
            depth=4,
            dim_head=64,
            heads=20,
            num_queries=num_tokens,
            embedding_dim=image_emb_dim,
            output_dim=self.unet.config.cross_attention_dim,
            ff_mult=4,
        )

        image_proj_model.eval()
        
        self.image_proj_model = image_proj_model.to(device, dtype=dtype)

        print('**************************** Loading image projection Model ***************************')
        if isinstance(model_ckpt, collections.OrderedDict):
            # print('Loading from state dict...')
            state_dict = model_ckpt
        elif isinstance(model_ckpt, str):
            # print(f'Loading state dict from {model_ckpt} ...')
            # state_dict = torch.load(model_ckpt, map_location="cpu", weights_only=True)
            state_dict = torch.load(model_ckpt, map_location="cpu", weights_only=True)
        else:
            raise TypeError("model_ckpt must be either an OrderedDict or a string (file path).")
        
        if isinstance(state_dict, tuple):
            print("\n\n\n state_dict is a tuple \n\n\n")
            state_dict = state_dict[0]

        self.image_proj_model.load_state_dict(state_dict)
        
        self.image_proj_model_in_features = image_emb_dim

        del state_dict
        gc.collect()

    def set_cross_attn(self, cross_attn_path, num_tokens):

        print('**************************** Setting cross attention processors to UNet ***************************')
        
        # self.unet    # 此时unet就是cnxs
        datatype = self.unet.dtype

        state_dict = torch.load(cross_attn_path, map_location="cpu", weights_only=True)
        attn_state_dict = {}
        for key, value in state_dict.items():
            if 'attn2.processor' in key:
                attn_state_dict[key] = value

        attn_procs = {}
        for name in self.unet.attn_processors.keys():
            if 'ctrl' in name:
                continue
            cross_attention_dim = None if name.endswith("attn1.processor") else self.unet.config.cross_attention_dim
            if name.startswith("mid_block"):
                hidden_size = self.unet.config.block_out_channels[-1]
            elif name.startswith("up_blocks"):
                block_id = int(name[len("up_blocks.")])
                hidden_size = list(reversed(self.unet.config.block_out_channels))[block_id]
            elif name.startswith("down_blocks"):
                block_id = int(name[len("down_blocks.")])
                hidden_size = self.unet.config.block_out_channels[block_id]
            
            if cross_attention_dim is None:
                attn_procs[name] = AttnProcessor()
            else:
                weights = {
                    "to_k_ip.weight": attn_state_dict[name + ".to_k_ip.weight"],
                    "to_v_ip.weight": attn_state_dict[name + ".to_v_ip.weight"],
                }
                attn_procs[name] = IPAttnProcessor(hidden_size=hidden_size, cross_attention_dim=cross_attention_dim, num_tokens=num_tokens)
                attn_procs[name].load_state_dict(weights)

        # print('length of attn_procs:', len(attn_procs)) # 140
        self.unet.set_attn_processor_unet(attn_procs)
        self.unet.to(dtype=datatype)

        del attn_state_dict
        del attn_procs
        gc.collect()

    def set_ip_adapter_scale(self, scale):
        unet = self.unet
        for attn_processor in unet.attn_processors_unet.values():
            # print(attn_processor)
            '''
            Attention(
                (to_q): Linear(in_features=640, out_features=640, bias=False)
                (to_k): Linear(in_features=2048, out_features=640, bias=False)
                (to_v): Linear(in_features=2048, out_features=640, bias=False)
                (to_out): ModuleList(
                    (0): Linear(in_features=640, out_features=640, bias=True)
                    (1): Dropout(p=0.0, inplace=False)
                )
                (processor): IPAttnProcessor2_0(
                    (to_k_ip): Linear(in_features=2048, out_features=640, bias=False)
                    (to_v_ip): Linear(in_features=2048, out_features=640, bias=False)
                )
                )
            '''
            if isinstance(attn_processor, IPAttnProcessor):
                # print('set_ip_adapter_scale: ',scale)
                attn_processor.scale = scale

    def _encode_prompt_image_emb(self, prompt_image_emb, device, num_images_per_prompt, dtype, do_classifier_free_guidance):
        
        if isinstance(prompt_image_emb, torch.Tensor):
            prompt_image_emb = prompt_image_emb.clone().detach()
        else:
            prompt_image_emb = torch.tensor(prompt_image_emb)
            
        prompt_image_emb = prompt_image_emb.reshape([1, -1, self.image_proj_model_in_features])
        
        if do_classifier_free_guidance:
            prompt_image_emb = torch.cat([torch.zeros_like(prompt_image_emb), prompt_image_emb], dim=0)
        else:
            prompt_image_emb = torch.cat([prompt_image_emb], dim=0)
        
        prompt_image_emb = prompt_image_emb.to(device=self.image_proj_model.latents.device, 
                                               dtype=self.image_proj_model.latents.dtype)
        prompt_image_emb = self.image_proj_model(prompt_image_emb)

        bs_embed, seq_len, _ = prompt_image_emb.shape
        prompt_image_emb = prompt_image_emb.repeat(1, num_images_per_prompt, 1)
        prompt_image_emb = prompt_image_emb.view(bs_embed * num_images_per_prompt, seq_len, -1)
        
        return prompt_image_emb.to(device=device, dtype=dtype)
    
    def load_lora_weights(
        self, pretrained_model_name_or_path_or_dict: Union[str, Dict[str, torch.Tensor]], adapter_name=None, **kwargs
    ):
        if not USE_PEFT_BACKEND:
            raise ValueError("PEFT backend is required for this method.")

        # if a dict is passed, copy it instead of modifying it inplace
        if isinstance(pretrained_model_name_or_path_or_dict, dict):
            pretrained_model_name_or_path_or_dict = pretrained_model_name_or_path_or_dict.copy()

        # First, ensure that the checkpoint is a compatible one and can be successfully loaded.
        if isinstance(pretrained_model_name_or_path_or_dict, str):
            filename = os.path.basename(pretrained_model_name_or_path_or_dict)
            extension = os.path.splitext(filename)[1]
            extension = extension[1:]
            if extension == "safetensors":
                lora_weight = load_file(pretrained_model_name_or_path_or_dict)
            else:
                lora_weight = torch.load(pretrained_model_name_or_path_or_dict, map_location="cpu")

            if all(
                (
                    k.startswith("lora_te_")
                    or k.startswith("lora_unet_")
                    or k.startswith("lora_te1_")
                    or k.startswith("lora_te2_")
                )
                for k in lora_weight.keys()
            ):
                state_dict = _maybe_map_sgm_blocks_to_diffusers(lora_weight, self.org_unet_config)
                state_dict, network_alphas = _convert_non_diffusers_lora_to_diffusers(state_dict)
                state_dict = get_module_kohya_state_dict_xs(state_dict, torch.float16)
                state_dict, _ = self.lora_state_dict(state_dict, **kwargs)
            else:
                state_dict = get_module_kohya_state_dict_xs(lora_weight, torch.float16)
                state_dict, network_alphas = self.lora_state_dict(state_dict, **kwargs)
        else:
            state_dict, network_alphas = self.lora_state_dict(pretrained_model_name_or_path_or_dict, **kwargs)


        is_correct_format = all("lora" in key for key in state_dict.keys())
        if not is_correct_format:
            raise ValueError("Invalid LoRA checkpoint.")
        
        low_cpu_mem_usage = False
        is_torch_higher_equal_2_1 = is_torch_version(">=", "2.1")
        
        if is_torch_higher_equal_2_1:
            from diffusers.models.modeling_utils import _LOW_CPU_MEM_USAGE_DEFAULT
            low_cpu_mem_usage = kwargs.pop("low_cpu_mem_usage", _LOW_CPU_MEM_USAGE_DEFAULT)
        
        assert is_torch_higher_equal_2_1 == low_cpu_mem_usage
        
        self.load_lora_into_unet(
            state_dict,
            network_alphas=network_alphas,
            unet=getattr(self, self.unet_name) if not hasattr(self, "unet") else self.unet,
            adapter_name=adapter_name,
            _pipeline=self,
            low_cpu_mem_usage=low_cpu_mem_usage,
        )
        self.load_lora_into_text_encoder(
            state_dict,
            network_alphas=network_alphas,
            text_encoder=getattr(self, self.text_encoder_name) if not hasattr(self, "text_encoder") else self.text_encoder,
            lora_scale=self.lora_scale,
            adapter_name=adapter_name,
            _pipeline=self,
            low_cpu_mem_usage=low_cpu_mem_usage,
        )
   
    def set_adapters(
        self,
        adapter_names: Union[List[str], str],
        adapter_weights: Optional[Union[List[float], float]] = None,
    ):
        """
        Set the currently active adapters for use in the UNet.

        Args:
            adapter_names (`List[str]` or `str`):
                The names of the adapters to use.
            adapter_weights (`Union[List[float], float]`, *optional*):
                The adapter(s) weights to use with the UNet. If `None`, the weights are set to `1.0` for all the
                adapters.

        Example:

        ```py
        from diffusers import AutoPipelineForText2Image
        import torch

        pipeline = AutoPipelineForText2Image.from_pretrained(
            "stabilityai/stable-diffusion-xl-base-1.0", torch_dtype=torch.float16
        ).to("cuda")
        pipeline.load_lora_weights(
            "jbilcke-hf/sdxl-cinematic-1", weight_name="pytorch_lora_weights.safetensors", adapter_name="cinematic"
        )
        pipeline.load_lora_weights("nerijs/pixel-art-xl", weight_name="pixel-art-xl.safetensors", adapter_name="pixel")
        pipeline.set_adapters(["cinematic", "pixel"], adapter_weights=[0.5, 0.5])
        ```
        """

        if not USE_PEFT_BACKEND:
            raise ValueError("PEFT backend is required for `set_adapters()`.")

        adapter_names = [adapter_names] if isinstance(adapter_names, str) else adapter_names

        if adapter_weights is None:
            adapter_weights = [1.0] * len(adapter_names)
        elif isinstance(adapter_weights, float):
            adapter_weights = [adapter_weights] * len(adapter_names)

        if len(adapter_names) != len(adapter_weights):
            raise ValueError(
                f"Length of adapter names {len(adapter_names)} is not equal to the length of their weights {len(adapter_weights)}."
            )

        set_weights_and_activate_adapters(self.unet, adapter_names, adapter_weights)
    
    '''
    def disable_lora(self):
        """
        Disable the UNet's active LoRA layers.

        Example:

        ```py
        from diffusers import AutoPipelineForText2Image
        import torch

        pipeline = AutoPipelineForText2Image.from_pretrained(
            "stabilityai/stable-diffusion-xl-base-1.0", torch_dtype=torch.float16
        ).to("cuda")
        pipeline.load_lora_weights(
            "jbilcke-hf/sdxl-cinematic-1", weight_name="pytorch_lora_weights.safetensors", adapter_name="cinematic"
        )
        pipeline.disable_lora()
        ```
        """
        if not USE_PEFT_BACKEND:
            raise ValueError("PEFT backend is required for this method.")
        set_adapter_layers(self.unet, enabled=False)

    def enable_lora(self):
        """
        Enable the UNet's active LoRA layers.

        Example:

        ```py
        from diffusers import AutoPipelineForText2Image
        import torch

        pipeline = AutoPipelineForText2Image.from_pretrained(
            "stabilityai/stable-diffusion-xl-base-1.0", torch_dtype=torch.float16
        ).to("cuda")
        pipeline.load_lora_weights(
            "jbilcke-hf/sdxl-cinematic-1", weight_name="pytorch_lora_weights.safetensors", adapter_name="cinematic"
        )
        pipeline.enable_lora()
        ```
        """
        if not USE_PEFT_BACKEND:
            raise ValueError("PEFT backend is required for this method.")
        set_adapter_layers(self.unet, enabled=True)

    def delete_adapters(self, adapter_names: Union[List[str], str]):
        """
        Delete an adapter's LoRA layers from the UNet.

        Args:
            adapter_names (`Union[List[str], str]`):
                The names (single string or list of strings) of the adapter to delete.

        Example:

        ```py
        from diffusers import AutoPipelineForText2Image
        import torch

        pipeline = AutoPipelineForText2Image.from_pretrained(
            "stabilityai/stable-diffusion-xl-base-1.0", torch_dtype=torch.float16
        ).to("cuda")
        pipeline.load_lora_weights(
            "jbilcke-hf/sdxl-cinematic-1", weight_name="pytorch_lora_weights.safetensors", adapter_names="cinematic"
        )
        pipeline.delete_adapters("cinematic")
        ```
        """
        if not USE_PEFT_BACKEND:
            raise ValueError("PEFT backend is required for this method.")

        if isinstance(adapter_names, str):
            adapter_names = [adapter_names]

        for adapter_name in adapter_names:
            delete_adapter_layers(self.unet, adapter_name)

            # Pop also the corresponding adapter from the config
            if hasattr(self.unet, "peft_config"):
                self.unet.peft_config.pop(adapter_name, None)
    '''
    
    @torch.no_grad()
    @replace_example_docstring(EXAMPLE_DOC_STRING)
    def __call__(
        self,
        prompt: Union[str, List[str]] = None,
        prompt_2: Optional[Union[str, List[str]]] = None,
        image: PipelineImageInput = None,
        face_emb: Optional[torch.Tensor] = None,
        height: Optional[int] = None,
        width: Optional[int] = None,
        num_inference_steps: int = 50,
        guidance_scale: float = 5.0,
        negative_prompt: Optional[Union[str, List[str]]] = None,
        negative_prompt_2: 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.Tensor] = None,
        prompt_embeds: Optional[torch.Tensor] = None,
        negative_prompt_embeds: Optional[torch.Tensor] = None,
        pooled_prompt_embeds: Optional[torch.Tensor] = None,
        negative_pooled_prompt_embeds: Optional[torch.Tensor] = None,
        output_type: Optional[str] = "pil",
        return_dict: bool = True,
        cross_attention_kwargs: Optional[Dict[str, Any]] = None,
        controlnet_conditioning_scale: Union[float, List[float]] = 1.0,
        control_guidance_start: float = 0.0,
        control_guidance_end: float = 1.0,
        original_size: Tuple[int, int] = None,
        crops_coords_top_left: Tuple[int, int] = (0, 0),
        target_size: Tuple[int, int] = None,
        negative_original_size: Optional[Tuple[int, int]] = None,
        negative_crops_coords_top_left: Tuple[int, int] = (0, 0),
        negative_target_size: Optional[Tuple[int, int]] = None,
        clip_skip: Optional[int] = None,
        callback_on_step_end: Optional[
            Union[Callable[[int, int, Dict], None], PipelineCallback, MultiPipelineCallbacks]
        ] = None,
        callback_on_step_end_tensor_inputs: List[str] = ["latents"],

        # IP adapter
        ip_adapter_scale=None,
    ):
        r"""
        The call function to the pipeline for generation.

        Args:
            prompt (`str` or `List[str]`, *optional*):
                The prompt or prompts to guide image generation. If not defined, you need to pass `prompt_embeds`.
            prompt_2 (`str` or `List[str]`, *optional*):
                The prompt or prompts to be sent to `tokenizer_2` and `text_encoder_2`. If not defined, `prompt` is
                used in both text-encoders.
            image (`torch.Tensor`, `PIL.Image.Image`, `np.ndarray`, `List[torch.Tensor]`, `List[PIL.Image.Image]`, `List[np.ndarray]`,:
                    `List[List[torch.Tensor]]`, `List[List[np.ndarray]]` or `List[List[PIL.Image.Image]]`):
                The ControlNet input condition to provide guidance to the `unet` for generation. If the type is
                specified as `torch.Tensor`, it is passed to ControlNet as is. `PIL.Image.Image` can also be accepted
                as an image. The dimensions of the output image defaults to `image`'s dimensions. If height and/or
                width are passed, `image` is resized accordingly. If multiple ControlNets are specified in `init`,
                images must be passed as a list such that each element of the list can be correctly batched for input
                to a single ControlNet.
            height (`int`, *optional*, defaults to `self.unet.config.sample_size * self.vae_scale_factor`):
                The height in pixels of the generated image. Anything below 512 pixels won't work well for
                [stabilityai/stable-diffusion-xl-base-1.0](https://huggingface.co/stabilityai/stable-diffusion-xl-base-1.0)
                and checkpoints that are not specifically fine-tuned on low resolutions.
            width (`int`, *optional*, defaults to `self.unet.config.sample_size * self.vae_scale_factor`):
                The width in pixels of the generated image. Anything below 512 pixels won't work well for
                [stabilityai/stable-diffusion-xl-base-1.0](https://huggingface.co/stabilityai/stable-diffusion-xl-base-1.0)
                and checkpoints that are not specifically fine-tuned on low resolutions.
            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 5.0):
                A higher guidance scale value encourages the model to generate images closely linked to the text
                `prompt` at the expense of lower image quality. Guidance scale is enabled when `guidance_scale > 1`.
            negative_prompt (`str` or `List[str]`, *optional*):
                The prompt or prompts to guide what to not include in image generation. If not defined, you need to
                pass `negative_prompt_embeds` instead. Ignored when not using guidance (`guidance_scale < 1`).
            negative_prompt_2 (`str` or `List[str]`, *optional*):
                The prompt or prompts to guide what to not include in image generation. This is sent to `tokenizer_2`
                and `text_encoder_2`. If not defined, `negative_prompt` is used in both text-encoders.
            num_images_per_prompt (`int`, *optional*, defaults to 1):
                The number of images to generate per prompt.
            eta (`float`, *optional*, defaults to 0.0):
                Corresponds to parameter eta (η) from the [DDIM](https://arxiv.org/abs/2010.02502) paper. Only applies
                to the [`~schedulers.DDIMScheduler`], and is ignored in other schedulers.
            generator (`torch.Generator` or `List[torch.Generator]`, *optional*):
                A [`torch.Generator`](https://pytorch.org/docs/stable/generated/torch.Generator.html) to make
                generation deterministic.
            latents (`torch.Tensor`, *optional*):
                Pre-generated noisy latents sampled from a Gaussian distribution, to be used as inputs for image
                generation. Can be used to tweak the same generation with different prompts. If not provided, a latents
                tensor is generated by sampling using the supplied random `generator`.
            prompt_embeds (`torch.Tensor`, *optional*):
                Pre-generated text embeddings. Can be used to easily tweak text inputs (prompt weighting). If not
                provided, text embeddings are generated from the `prompt` input argument.
            negative_prompt_embeds (`torch.Tensor`, *optional*):
                Pre-generated negative text embeddings. Can be used to easily tweak text inputs (prompt weighting). If
                not provided, `negative_prompt_embeds` are generated from the `negative_prompt` input argument.
            pooled_prompt_embeds (`torch.Tensor`, *optional*):
                Pre-generated pooled text embeddings. Can be used to easily tweak text inputs (prompt weighting). If
                not provided, pooled text embeddings are generated from `prompt` input argument.
            negative_pooled_prompt_embeds (`torch.Tensor`, *optional*):
                Pre-generated negative pooled text embeddings. Can be used to easily tweak text inputs (prompt
                weighting). If not provided, pooled `negative_prompt_embeds` are generated from `negative_prompt` input
                argument.
            output_type (`str`, *optional*, defaults to `"pil"`):
                The output format of the generated image. Choose between `PIL.Image` or `np.array`.
            return_dict (`bool`, *optional*, defaults to `True`):
                Whether or not to return a [`~pipelines.stable_diffusion.StableDiffusionPipelineOutput`] instead of a
                plain tuple.
            cross_attention_kwargs (`dict`, *optional*):
                A kwargs dictionary that if specified is passed along to the [`AttentionProcessor`] as defined in
                [`self.processor`](https://github.com/huggingface/diffusers/blob/main/src/diffusers/models/attention_processor.py).
            controlnet_conditioning_scale (`float` or `List[float]`, *optional*, defaults to 1.0):
                The outputs of the ControlNet are multiplied by `controlnet_conditioning_scale` before they are added
                to the residual in the original `unet`.
            control_guidance_start (`float`, *optional*, defaults to 0.0):
                The percentage of total steps at which the ControlNet starts applying.
            control_guidance_end (`float`, *optional*, defaults to 1.0):
                The percentage of total steps at which the ControlNet stops applying.
            original_size (`Tuple[int]`, *optional*, defaults to (1024, 1024)):
                If `original_size` is not the same as `target_size` the image will appear to be down- or upsampled.
                `original_size` defaults to `(width, height)` if not specified. Part of SDXL's micro-conditioning as
                explained in section 2.2 of
                [https://huggingface.co/papers/2307.01952](https://huggingface.co/papers/2307.01952).
            crops_coords_top_left (`Tuple[int]`, *optional*, defaults to (0, 0)):
                `crops_coords_top_left` can be used to generate an image that appears to be "cropped" from the position
                `crops_coords_top_left` downwards. Favorable, well-centered images are usually achieved by setting
                `crops_coords_top_left` to (0, 0). Part of SDXL's micro-conditioning as explained in section 2.2 of
                [https://huggingface.co/papers/2307.01952](https://huggingface.co/papers/2307.01952).
            target_size (`Tuple[int]`, *optional*, defaults to (1024, 1024)):
                For most cases, `target_size` should be set to the desired height and width of the generated image. If
                not specified it will default to `(width, height)`. Part of SDXL's micro-conditioning as explained in
                section 2.2 of [https://huggingface.co/papers/2307.01952](https://huggingface.co/papers/2307.01952).
            negative_original_size (`Tuple[int]`, *optional*, defaults to (1024, 1024)):
                To negatively condition the generation process based on a specific image resolution. Part of SDXL's
                micro-conditioning as explained in section 2.2 of
                [https://huggingface.co/papers/2307.01952](https://huggingface.co/papers/2307.01952). For more
                information, refer to this issue thread: https://github.com/huggingface/diffusers/issues/4208.
            negative_crops_coords_top_left (`Tuple[int]`, *optional*, defaults to (0, 0)):
                To negatively condition the generation process based on a specific crop coordinates. Part of SDXL's
                micro-conditioning as explained in section 2.2 of
                [https://huggingface.co/papers/2307.01952](https://huggingface.co/papers/2307.01952). For more
                information, refer to this issue thread: https://github.com/huggingface/diffusers/issues/4208.
            negative_target_size (`Tuple[int]`, *optional*, defaults to (1024, 1024)):
                To negatively condition the generation process based on a target image resolution. It should be as same
                as the `target_size` for most cases. Part of SDXL's micro-conditioning as explained in section 2.2 of
                [https://huggingface.co/papers/2307.01952](https://huggingface.co/papers/2307.01952). For more
                information, refer to this issue thread: https://github.com/huggingface/diffusers/issues/4208.
            clip_skip (`int`, *optional*):
                Number of layers to be skipped from CLIP while computing the prompt embeddings. A value of 1 means that
                the output of the pre-final layer will be used for computing the prompt embeddings.
            callback_on_step_end (`Callable`, `PipelineCallback`, `MultiPipelineCallbacks`, *optional*):
                A function or a subclass of `PipelineCallback` or `MultiPipelineCallbacks` that is called at the end of
                each denoising step during the inference. with the following arguments: `callback_on_step_end(self:
                DiffusionPipeline, step: int, timestep: int, callback_kwargs: Dict)`. `callback_kwargs` will include a
                list of all tensors as specified by `callback_on_step_end_tensor_inputs`.
            callback_on_step_end_tensor_inputs (`List`, *optional*):
                The list of tensor inputs for the `callback_on_step_end` function. The tensors specified in the list
                will be passed as `callback_kwargs` argument. You will only be able to include variables listed in the
                `._callback_tensor_inputs` attribute of your pipeine class.

        Examples:

        Returns:
            [`~pipelines.stable_diffusion.StableDiffusionXLPipelineOutput`] or `tuple`:
                If `return_dict` is `True`, [`~pipelines.stable_diffusion.StableDiffusionXLPipelineOutput`] is
                returned, otherwise a `tuple` is returned containing the output images.
        """
        
        lpw = LongPromptWeight()
        
        if isinstance(callback_on_step_end, (PipelineCallback, MultiPipelineCallbacks)):
            callback_on_step_end_tensor_inputs = callback_on_step_end.tensor_inputs

        unet = self.unet._orig_mod if is_compiled_module(self.unet) else self.unet
        
        # 0. set ip_adapter_scale
        if ip_adapter_scale is not None:
            self.set_ip_adapter_scale(ip_adapter_scale)

        # 1. Check inputs. Raise error if not correct
        self.check_inputs(
            prompt,
            prompt_2,
            image,
            negative_prompt,
            negative_prompt_2,
            prompt_embeds,
            negative_prompt_embeds,
            pooled_prompt_embeds,
            negative_pooled_prompt_embeds,
            controlnet_conditioning_scale,
            control_guidance_start,
            control_guidance_end,
            callback_on_step_end_tensor_inputs,
        )

        self._guidance_scale = guidance_scale
        self._clip_skip = clip_skip
        self._cross_attention_kwargs = cross_attention_kwargs
        self._interrupt = False

        # 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

        # 3. Encode input prompt
        
        # text_encoder_lora_scale = (
        #     self.cross_attention_kwargs.get("scale", None) if self.cross_attention_kwargs is not None else None
        # )
        # (
        #     prompt_embeds,
        #     negative_prompt_embeds,
        #     pooled_prompt_embeds,
        #     negative_pooled_prompt_embeds,
        # ) = self.encode_prompt(
        #     prompt,
        #     prompt_2,
        #     device,
        #     num_images_per_prompt,
        #     do_classifier_free_guidance,
        #     negative_prompt,
        #     negative_prompt_2,
        #     prompt_embeds=prompt_embeds,
        #     negative_prompt_embeds=negative_prompt_embeds,
        #     pooled_prompt_embeds=pooled_prompt_embeds,
        #     negative_pooled_prompt_embeds=negative_pooled_prompt_embeds,
        #     lora_scale=text_encoder_lora_scale,
        #     clip_skip=clip_skip,
        # )

        (
            prompt_embeds,
            negative_prompt_embeds,
            pooled_prompt_embeds,
            negative_pooled_prompt_embeds,
        ) = lpw.get_weighted_text_embeddings_sdxl(
            pipe=self,
            prompt=prompt,
            neg_prompt=negative_prompt,
            prompt_embeds=prompt_embeds,
            negative_prompt_embeds=negative_prompt_embeds,
            pooled_prompt_embeds=pooled_prompt_embeds,
            negative_pooled_prompt_embeds=negative_pooled_prompt_embeds,
        )
        prompt_image_emb = self._encode_prompt_image_emb(
            face_emb, 
            device,
            num_images_per_prompt,
            unet.dtype,
            do_classifier_free_guidance
        )

        # 4. Prepare image
        if isinstance(unet, UNetControlNetXSModel):
            image = self.prepare_image(
                image=image,
                width=width,
                height=height,
                batch_size=batch_size * num_images_per_prompt,
                num_images_per_prompt=num_images_per_prompt,
                device=device,
                dtype=unet.dtype,
                do_classifier_free_guidance=do_classifier_free_guidance,
            )
            height, width = image.shape[-2:]
        else:
            assert False

        # 5. Prepare timesteps
        self.scheduler.set_timesteps(num_inference_steps, device=device)
        timesteps = self.scheduler.timesteps

        # 6. 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,
            prompt_embeds.dtype,
            device,
            generator,
            latents,
        )

        # 7. Prepare extra step kwargs. TODO: Logic should ideally just be moved out of the pipeline
        extra_step_kwargs = self.prepare_extra_step_kwargs(generator, eta)

        # 7.1 Prepare added time ids & embeddings
        if isinstance(image, list):
            original_size = original_size or image[0].shape[-2:]
        else:
            original_size = original_size or image.shape[-2:]
        target_size = target_size or (height, width)

        add_text_embeds = pooled_prompt_embeds
        if self.text_encoder_2 is None:
            text_encoder_projection_dim = int(pooled_prompt_embeds.shape[-1])
        else:
            text_encoder_projection_dim = self.text_encoder_2.config.projection_dim

        add_time_ids = self._get_add_time_ids(
            original_size,
            crops_coords_top_left,
            target_size,
            dtype=prompt_embeds.dtype,
            text_encoder_projection_dim=text_encoder_projection_dim,
        )

        if negative_original_size is not None and negative_target_size is not None:
            negative_add_time_ids = self._get_add_time_ids(
                negative_original_size,
                negative_crops_coords_top_left,
                negative_target_size,
                dtype=prompt_embeds.dtype,
                text_encoder_projection_dim=text_encoder_projection_dim,
            )
        else:
            negative_add_time_ids = add_time_ids

        if do_classifier_free_guidance:
            prompt_embeds = torch.cat([negative_prompt_embeds, prompt_embeds], dim=0)
            add_text_embeds = torch.cat([negative_pooled_prompt_embeds, add_text_embeds], dim=0)
            add_time_ids = torch.cat([negative_add_time_ids, add_time_ids], dim=0)

        prompt_embeds = prompt_embeds.to(device, dtype=unet.dtype)
        add_text_embeds = add_text_embeds.to(device, dtype=unet.dtype)
        add_time_ids = add_time_ids.to(device, dtype=unet.dtype).repeat(batch_size * num_images_per_prompt, 1)
        
        prompt_image_emb = prompt_image_emb.to(device, dtype=unet.dtype)
        encoder_hidden_states = torch.cat([prompt_embeds, prompt_image_emb], dim=1)
        encoder_hidden_states = encoder_hidden_states.to(device, dtype=unet.dtype)
        

        # 8. Denoising loop
        num_warmup_steps = len(timesteps) - num_inference_steps * self.scheduler.order
        self._num_timesteps = len(timesteps)
        is_controlnet_compiled = is_compiled_module(self.unet)
        is_torch_higher_equal_2_1 = is_torch_version(">=", "2.1")

        with self.progress_bar(total=num_inference_steps) as progress_bar:
            for i, t in enumerate(timesteps):
                # Relevant thread:
                # https://dev-discuss.pytorch.org/t/cudagraphs-in-pytorch-2-0/1428
                if is_controlnet_compiled and is_torch_higher_equal_2_1:
                    torch._inductor.cudagraph_mark_step_begin()
                # expand the latents if we are doing classifier free guidance
                latent_model_input = torch.cat([latents] * 2) if do_classifier_free_guidance else latents
                latent_model_input = self.scheduler.scale_model_input(latent_model_input, t)

                added_cond_kwargs = {"text_embeds": add_text_embeds, "time_ids": add_time_ids}

                # predict the noise residual
                apply_control = (
                    i / len(timesteps) >= control_guidance_start and (i + 1) / len(timesteps) <= control_guidance_end
                )

                noise_pred = self.unet(
                    sample=latent_model_input,
                    timestep=t,
                    unet_encoder_hidden_states=encoder_hidden_states, 
                    cnxs_encoder_hidden_states=prompt_image_emb, 
                    controlnet_cond=image,
                    conditioning_scale=controlnet_conditioning_scale,
                    cross_attention_kwargs=self.cross_attention_kwargs,
                    added_cond_kwargs=added_cond_kwargs,
                    return_dict=True,
                    apply_control=apply_control,
                ).sample

                # perform guidance
                if do_classifier_free_guidance:
                    noise_pred_uncond, noise_pred_text = noise_pred.chunk(2)
                    noise_pred = noise_pred_uncond + guidance_scale * (noise_pred_text - noise_pred_uncond)

                # compute the previous noisy sample x_t -> x_t-1
                latents = self.scheduler.step(noise_pred, t, latents, **extra_step_kwargs, return_dict=False)[0]

                if callback_on_step_end is not None:
                    callback_kwargs = {}
                    for k in callback_on_step_end_tensor_inputs:
                        callback_kwargs[k] = locals()[k]
                    callback_outputs = callback_on_step_end(self, i, t, callback_kwargs)

                    latents = callback_outputs.pop("latents", latents)
                    prompt_embeds = callback_outputs.pop("prompt_embeds", prompt_embeds)
                    negative_prompt_embeds = callback_outputs.pop("negative_prompt_embeds", negative_prompt_embeds)

                # call the callback, if provided
                if i == len(timesteps) - 1 or ((i + 1) > num_warmup_steps and (i + 1) % self.scheduler.order == 0):
                    progress_bar.update()

        # manually for max memory savings
        if self.vae.dtype == torch.float16 and self.vae.config.force_upcast:
            self.upcast_vae()
            latents = latents.to(next(iter(self.vae.post_quant_conv.parameters())).dtype)

        if not output_type == "latent":
            # make sure the VAE is in float32 mode, as it overflows in float16
            needs_upcasting = self.vae.dtype == torch.float16 and self.vae.config.force_upcast

            if needs_upcasting:
                self.upcast_vae()
                latents = latents.to(next(iter(self.vae.post_quant_conv.parameters())).dtype)

            image = self.vae.decode(latents / self.vae.config.scaling_factor, return_dict=False)[0]

            # cast back to fp16 if needed
            if needs_upcasting:
                self.vae.to(dtype=torch.float16)
        else:
            image = latents

        if not output_type == "latent":
            # apply watermark if available
            if self.watermark is not None:
                image = self.watermark.apply_watermark(image)

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

        # Offload all models
        self.maybe_free_model_hooks()

        if not return_dict:
            return (image,)

        return StableDiffusionXLPipelineOutput(images=image)