Upload folder using huggingface_hub

inference.py

@@ -15,7 +15,8 @@ from internals.util.args import apply_style_args
 from internals.util.avatar import Avatar
 from internals.util.cache import auto_clear_cuda_and_gc, clear_cuda
 from internals.util.commons import pickPoses, upload_image, upload_images
-from internals.util.config import set_configs_from_task, set_root_dir
+from internals.util.config import (num_return_sequences, set_configs_from_task,
+                                   set_root_dir)
 from internals.util.failure_hander import FailureHandler
 from internals.util.lora_style import LoraStyle
 from internals.util.slack import Slack
@@ -23,7 +24,6 @@ from internals.util.slack import Slack
 torch.backends.cudnn.benchmark = True
 torch.backends.cuda.matmul.allow_tf32 = True
 
-num_return_sequences = 4  # the number of results to generate
 auto_mode = False
 
 prompt_modifier = PromptModifier(num_of_sequences=num_return_sequences)
@@ -57,10 +57,6 @@ def get_patched_prompt(task: Task):
     ori_prompt = [task.get_prompt()] * num_return_sequences
 
     class_name = None
-    # if task.get_imageUrl():
-    #     class_name = img_classifier.classify(
-    #         task.get_imageUrl(), task.get_width(), task.get_height()
-    #     )
     add_style_and_character(ori_prompt, class_name)
     add_style_and_character(prompt, class_name)
 
@@ -69,6 +65,54 @@ def get_patched_prompt(task: Task):
     return (prompt, ori_prompt)
 
 
+def get_patched_prompt_text2img(task: Task) -> Text2Img.Params:
+    def add_style_and_character(prompt: str, prepend: str = ""):
+        prompt = avatar.add_code_names(prompt)
+        prompt = lora_style.prepend_style_to_prompt(prompt, task.get_style())
+        prompt = prepend + prompt
+        return prompt
+
+    if task.get_prompt_left() and task.get_prompt_right():
+        # prepend = "2characters, "
+        prepend = ""
+        if task.is_prompt_engineering():
+            mod_prompt = prompt_modifier.modify(task.get_prompt())
+        else:
+            mod_prompt = [task.get_prompt()] * num_return_sequences
+
+        prompt, prompt_left, prompt_right = [], [], []
+        for i in range(len(mod_prompt)):
+            mp = mod_prompt[i].replace(task.get_prompt(), "")
+            prompt.append(add_style_and_character(task.get_prompt(), prepend) + mp)
+            prompt_left.append(
+                add_style_and_character(task.get_prompt_left(), prepend) + mp
+            )
+            prompt_right.append(
+                add_style_and_character(task.get_prompt_right(), prepend) + mp
+            )
+
+        params = Text2Img.Params(
+            prompt=prompt,
+            prompt_left=prompt_left,
+            prompt_right=prompt_right,
+        )
+    else:
+        if task.is_prompt_engineering():
+            mod_prompt = prompt_modifier.modify(task.get_prompt())
+        else:
+            mod_prompt = [task.get_prompt()] * num_return_sequences
+        mod_prompt = [add_style_and_character(mp) for mp in mod_prompt]
+
+        params = Text2Img.Params(
+            prompt=[add_style_and_character(task.get_prompt())] * num_return_sequences,
+            modified_prompt=mod_prompt,
+        )
+
+    print(params)
+
+    return params
+
+
 def get_patched_prompt_tile_upscale(task: Task):
     if task.get_prompt():
         prompt = task.get_prompt()
@@ -164,6 +208,72 @@ def tile_upscale(task: Task):
     }
 
 
+@update_db
+@auto_clear_cuda_and_gc(controlnet)
+@slack.auto_send_alert
+def scribble(task: Task):
+    prompt, _ = get_patched_prompt(task)
+
+    controlnet.load_scribble()
+
+    lora_patcher = lora_style.get_patcher(controlnet.pipe2, task.get_style())
+    lora_patcher.patch()
+
+    images, has_nsfw = controlnet.process_scribble(
+        imageUrl=task.get_imageUrl(),
+        seed=task.get_seed(),
+        steps=task.get_steps(),
+        width=task.get_width(),
+        height=task.get_height(),
+        prompt=prompt,
+        negative_prompt=[task.get_negative_prompt()] * num_return_sequences,
+    )
+
+    generated_image_urls = upload_images(images, "_scribble", task.get_taskId())
+
+    lora_patcher.cleanup()
+    controlnet.cleanup()
+
+    return {
+        "modified_prompts": prompt,
+        "generated_image_urls": generated_image_urls,
+        "has_nsfw": has_nsfw,
+    }
+
+
+@update_db
+@auto_clear_cuda_and_gc(controlnet)
+@slack.auto_send_alert
+def linearart(task: Task):
+    prompt, _ = get_patched_prompt(task)
+
+    controlnet.load_linearart()
+
+    lora_patcher = lora_style.get_patcher(controlnet.pipe2, task.get_style())
+    lora_patcher.patch()
+
+    images, has_nsfw = controlnet.process_linearart(
+        imageUrl=task.get_imageUrl(),
+        seed=task.get_seed(),
+        steps=task.get_steps(),
+        width=task.get_width(),
+        height=task.get_height(),
+        prompt=prompt,
+        negative_prompt=[task.get_negative_prompt()] * num_return_sequences,
+    )
+
+    generated_image_urls = upload_images(images, "_linearart", task.get_taskId())
+
+    lora_patcher.cleanup()
+    controlnet.cleanup()
+
+    return {
+        "modified_prompts": prompt,
+        "generated_image_urls": generated_image_urls,
+        "has_nsfw": has_nsfw,
+    }
+
+
 @update_db
 @auto_clear_cuda_and_gc(controlnet)
 @slack.auto_send_alert
@@ -207,7 +317,7 @@ def pose(task: Task, s3_outkey: str = "_pose", poses: Optional[list] = None):
 @auto_clear_cuda_and_gc(controlnet)
 @slack.auto_send_alert
 def text2img(task: Task):
-    prompt, ori_prompt = get_patched_prompt(task)
+    params = get_patched_prompt_text2img(task)
 
     lora_patcher = lora_style.get_patcher(text2img_pipe.pipe, task.get_style())
     lora_patcher.patch()
@@ -215,13 +325,12 @@ def text2img(task: Task):
     torch.manual_seed(task.get_seed())
 
     images, has_nsfw = text2img_pipe.process(
-        prompt=ori_prompt,
-        modified_prompts=prompt,
+        params=params,
         num_inference_steps=task.get_steps(),
         guidance_scale=7.5,
         height=task.get_height(),
         width=task.get_width(),
-        negative_prompt=[task.get_negative_prompt()] * num_return_sequences,
+        negative_prompt=task.get_negative_prompt(),
         iteration=task.get_iteration(),
         **lora_patcher.kwargs(),
     )
@@ -231,7 +340,7 @@ def text2img(task: Task):
     lora_patcher.cleanup()
 
     return {
-        "modified_prompts": prompt,
+        **params.__dict__,
         "generated_image_urls": generated_image_urls,
         "has_nsfw": has_nsfw,
     }
@@ -361,6 +470,10 @@ def predict_fn(data, pipe):
             return tile_upscale(task)
         elif task_type == TaskType.INPAINT:
             return inpaint(task)
+        elif task_type == TaskType.SCRIBBLE:
+            return scribble(task)
+        elif task_type == TaskType.LINEARART:
+            return linearart(task)
         else:
             raise Exception("Invalid task type")
     except Exception as e:

inference2.py

@@ -12,16 +12,18 @@ from internals.pipelines.safety_checker import SafetyChecker
 from internals.pipelines.upscaler import Upscaler
 from internals.util.avatar import Avatar
 from internals.util.cache import clear_cuda
-from internals.util.commons import (construct_default_s3_url, upload_image,
-                                    upload_images)
-from internals.util.config import set_configs_from_task, set_root_dir
+from internals.util.commons import construct_default_s3_url, upload_image, upload_images
+from internals.util.config import (
+    num_return_sequences,
+    set_configs_from_task,
+    set_root_dir,
+)
 from internals.util.failure_hander import FailureHandler
 from internals.util.slack import Slack
 
 torch.backends.cudnn.benchmark = True
 torch.backends.cuda.matmul.allow_tf32 = True
 
-num_return_sequences = 4
 auto_mode = False
 
 slack = Slack()

internals/data/task.py

@@ -14,6 +14,8 @@ class TaskType(Enum):
     UPSCALE_IMAGE = "UPSCALE_IMAGE"
     TILE_UPSCALE = "TILE_UPSCALE"
     OBJECT_REMOVAL = "OBJECT_REMOVAL"
+    SCRIBBLE = "SCRIBBLE"
+    LINEARART = "LINEARART"
 
 
 class ModelType(Enum):
@@ -45,6 +47,12 @@ class Task:
     def get_prompt(self) -> str:
         return self.__data.get("prompt", "")
 
+    def get_prompt_left(self) -> str:
+        return self.__data.get("prompt_left", "")
+
+    def get_prompt_right(self) -> str:
+        return self.__data.get("prompt_right", "")
+
     def get_userId(self) -> str:
         return self.__data.get("userId", "")
 

internals/pipelines/commons.py

@@ -1,3 +1,4 @@
+from dataclasses import dataclass
 from typing import Any, Callable, Dict, List, Optional, Union
 
 import torch
@@ -6,6 +7,7 @@ from diffusers import StableDiffusionImg2ImgPipeline
 from internals.data.result import Result
 from internals.pipelines.twoStepPipeline import two_step_pipeline
 from internals.util.commons import disable_safety_checker, download_image
+from internals.util.config import num_return_sequences
 
 
 class AbstractPipeline:
@@ -17,6 +19,13 @@ class AbstractPipeline:
 
 
 class Text2Img(AbstractPipeline):
+    @dataclass
+    class Params:
+        prompt: List[str] = None
+        modified_prompt: List[str] = None
+        prompt_left: List[str] = None
+        prompt_right: List[str] = None
+
     def load(self, model_dir: str):
         self.pipe = two_step_pipeline.from_pretrained(
             model_dir, torch_dtype=torch.float16
@@ -33,14 +42,13 @@ class Text2Img(AbstractPipeline):
     @torch.inference_mode()
     def process(
         self,
-        prompt: Union[str, List[str]] = None,
-        modified_prompts: Union[str, List[str]] = None,
+        params: Params,
         height: Optional[int] = None,
         width: Optional[int] = None,
         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,
+        negative_prompt: Optional[str] = None,
+        num_images_per_prompt: int = 1,
         eta: float = 0.0,
         generator: Optional[Union[torch.Generator, List[torch.Generator]]] = None,
         latents: Optional[torch.FloatTensor] = None,
@@ -53,27 +61,54 @@ class Text2Img(AbstractPipeline):
         cross_attention_kwargs: Optional[Dict[str, Any]] = None,
         iteration: float = 3.0,
     ):
-        result = self.pipe.two_step_pipeline(
-            prompt=prompt,
-            modified_prompts=modified_prompts,
-            height=height,
-            width=width,
-            num_inference_steps=num_inference_steps,
-            guidance_scale=guidance_scale,
-            negative_prompt=negative_prompt,
-            num_images_per_prompt=num_images_per_prompt,
-            eta=eta,
-            generator=generator,
-            latents=latents,
-            prompt_embeds=prompt_embeds,
-            negative_prompt_embeds=negative_prompt_embeds,
-            output_type=output_type,
-            return_dict=return_dict,
-            callback=callback,
-            callback_steps=callback_steps,
-            cross_attention_kwargs=cross_attention_kwargs,
-            iteration=iteration,
-        )
+        prompt = params.prompt
+
+        if params.prompt_left and params.prompt_right:
+            # multi-character pipelines
+            prompt = [params.prompt[0], params.prompt_left[0], params.prompt_right[0]]
+            result = self.pipe.multi_character_diffusion(
+                prompt=prompt,
+                pos=["1:1-0:0", "1:2-0:0", "1:2-0:1"],
+                mix_val=[0.2, 0.8, 0.8],
+                height=height,
+                width=width,
+                num_inference_steps=num_inference_steps,
+                guidance_scale=guidance_scale,
+                negative_prompt=[negative_prompt or ""] * len(prompt),
+                num_images_per_prompt=num_return_sequences,
+                eta=eta,
+                # generator=generator,
+                output_type=output_type,
+                return_dict=return_dict,
+                callback=callback,
+                callback_steps=callback_steps,
+            )
+        else:
+            # two step pipeline
+            modified_prompt = params.modified_prompt
+
+            result = self.pipe.two_step_pipeline(
+                prompt=prompt,
+                modified_prompts=modified_prompt,
+                height=height,
+                width=width,
+                num_inference_steps=num_inference_steps,
+                guidance_scale=guidance_scale,
+                negative_prompt=[negative_prompt or ""] * num_return_sequences,
+                num_images_per_prompt=num_images_per_prompt,
+                eta=eta,
+                generator=generator,
+                latents=latents,
+                prompt_embeds=prompt_embeds,
+                negative_prompt_embeds=negative_prompt_embeds,
+                output_type=output_type,
+                return_dict=return_dict,
+                callback=callback,
+                callback_steps=callback_steps,
+                cross_attention_kwargs=cross_attention_kwargs,
+                iteration=iteration,
+            )
+
         return Result.from_result(result)
 
 

internals/pipelines/controlnets.py

@@ -1,20 +1,20 @@
-from typing import List
+from typing import List, Union
 
 import cv2
 import numpy as np
 import torch
-from controlnet_aux import OpenposeDetector
-from diffusers import (
-    ControlNetModel,
-    DiffusionPipeline,
-    StableDiffusionControlNetPipeline,
-    UniPCMultistepScheduler,
-)
+from controlnet_aux import HEDdetector, LineartDetector, OpenposeDetector
+from diffusers import (ControlNetModel, DiffusionPipeline,
+                       StableDiffusionControlNetPipeline,
+                       UniPCMultistepScheduler)
 from PIL import Image
+from torch.nn import Linear
 from tqdm import gui
 
 from internals.data.result import Result
 from internals.pipelines.commons import AbstractPipeline
+from internals.pipelines.tileUpscalePipeline import \
+    StableDiffusionControlNetImg2ImgPipeline
 from internals.util.cache import clear_cuda_and_gc
 from internals.util.commons import download_image
 
@@ -27,11 +27,10 @@ class ControlNet(AbstractPipeline):
         self.load_canny()
 
         # controlnet pipeline for canny and pose
-        pipe = DiffusionPipeline.from_pretrained(
+        pipe = StableDiffusionControlNetImg2ImgPipeline.from_pretrained(
             model_dir,
             controlnet=self.controlnet,
             torch_dtype=torch.float16,
-            custom_pipeline="stable_diffusion_controlnet_img2img",
         ).to("cuda")
         pipe.scheduler = UniPCMultistepScheduler.from_config(pipe.scheduler.config)
         pipe.enable_model_cpu_offload()
@@ -62,7 +61,7 @@ class ControlNet(AbstractPipeline):
         if self.__current_task_name == "pose":
             return
         pose = ControlNetModel.from_pretrained(
-            "lllyasviel/control_v11p_sd15_openpose", torch_dtype=torch.float16
+            "lllyasviel/sd-controlnet-openpose", torch_dtype=torch.float16
         ).to("cuda")
         self.__current_task_name = "pose"
         self.controlnet = pose
@@ -86,6 +85,35 @@ class ControlNet(AbstractPipeline):
             self.pipe2.controlnet = tile_upscaler
         clear_cuda_and_gc()
 
+    def load_scribble(self):
+        if self.__current_task_name == "scribble":
+            return
+        scribble = ControlNetModel.from_pretrained(
+            "lllyasviel/control_v11p_sd15_scribble", torch_dtype=torch.float16
+        ).to("cuda")
+        self.__current_task_name = "scribble"
+        self.controlnet = scribble
+        if hasattr(self, "pipe"):
+            self.pipe.controlnet = scribble
+        if hasattr(self, "pipe2"):
+            self.pipe2.controlnet = scribble
+        clear_cuda_and_gc()
+
+    def load_linearart(self):
+        if self.__current_task_name == "linearart":
+            return
+        linearart = ControlNetModel.from_pretrained(
+            "ControlNet-1-1-preview/control_v11p_sd15_lineart",
+            torch_dtype=torch.float16,
+        ).to("cuda")
+        self.__current_task_name = "linearart"
+        self.controlnet = linearart
+        if hasattr(self, "pipe"):
+            self.pipe.controlnet = linearart
+        if hasattr(self, "pipe2"):
+            self.pipe2.controlnet = linearart
+        clear_cuda_and_gc()
+
     def cleanup(self):
         self.pipe.controlnet = None
         self.pipe2.controlnet = None
@@ -191,12 +219,84 @@ class ControlNet(AbstractPipeline):
         )
         return Result.from_result(result)
 
+    @torch.inference_mode()
+    def process_scribble(
+        self,
+        imageUrl: str,
+        prompt: Union[str, List[str]],
+        negative_prompt: Union[str, List[str]],
+        steps: int,
+        seed: int,
+        height: int,
+        width: int,
+        guidance_scale: float = 7.5,
+    ):
+        if self.__current_task_name != "scribble":
+            raise Exception("ControlNet is not loaded with scribble model")
+
+        torch.manual_seed(seed)
+
+        init_image = download_image(imageUrl).resize((width, height))
+        condition_image = self.__scribble_condition_image(init_image)
+
+        result = self.pipe2.__call__(
+            image=condition_image,
+            prompt=prompt,
+            num_inference_steps=steps,
+            negative_prompt=negative_prompt,
+            height=height,
+            width=width,
+            guidance_scale=guidance_scale,
+        )
+        return Result.from_result(result)
+
+    @torch.inference_mode()
+    def process_linearart(
+        self,
+        imageUrl: str,
+        prompt: Union[str, List[str]],
+        negative_prompt: Union[str, List[str]],
+        steps: int,
+        seed: int,
+        height: int,
+        width: int,
+        guidance_scale: float = 7.5,
+    ):
+        if self.__current_task_name != "linearart":
+            raise Exception("ControlNet is not loaded with linearart model")
+
+        torch.manual_seed(seed)
+
+        init_image = download_image(imageUrl).resize((width, height))
+        condition_image = self.__linearart_condition_image(init_image)
+
+        result = self.pipe2.__call__(
+            image=condition_image,
+            prompt=prompt,
+            num_inference_steps=steps,
+            negative_prompt=negative_prompt,
+            height=height,
+            width=width,
+            guidance_scale=guidance_scale,
+        )
+        return Result.from_result(result)
+
     def detect_pose(self, imageUrl: str) -> Image.Image:
         detector = OpenposeDetector.from_pretrained("lllyasviel/ControlNet")
         image = download_image(imageUrl)
         image = detector.__call__(image, hand_and_face=True)
         return image
 
+    def __scribble_condition_image(self, image: Image.Image) -> Image.Image:
+        processor = HEDdetector.from_pretrained("lllyasviel/Annotators")
+        image = processor.__call__(input_image=image, scribble=True)
+        return image
+
+    def __linearart_condition_image(self, image: Image.Image) -> Image.Image:
+        processor = LineartDetector.from_pretrained("lllyasviel/Annotators")
+        image = processor.__call__(input_image=image)
+        return image
+
     def __canny_detect_edge(self, image: Image.Image) -> Image.Image:
         image_array = np.array(image)
 

internals/pipelines/prompt_modifier.py

@@ -18,18 +18,20 @@ class PromptModifier:
         self.prompter_tokenizer.pad_token = self.prompter_tokenizer.eos_token
         self.prompter_tokenizer.padding_side = "left"
 
-    def modify(self, text: str) -> List[str]:
+    def modify(self, text: str, num_of_sequences: Optional[int] = None) -> List[str]:
         eos_id = self.prompter_tokenizer.eos_token_id
         # restricted_words_list = ["octane", "cyber"]
         # restricted_words_token_ids = prompter_tokenizer(
         #     restricted_words_list, add_special_tokens=False
         # ).input_ids
 
+        num_of_sequences = num_of_sequences or self.__num_of_sequences
+
         generation_config = GenerationConfig(
             do_sample=False,
             max_new_tokens=75,
             num_beams=4,
-            num_return_sequences=self.__num_of_sequences,
+            num_return_sequences=num_of_sequences,
             eos_token_id=eos_id,
             pad_token_id=eos_id,
             length_penalty=-1.0,

internals/pipelines/tileUpscalePipeline.py

@@ -0,0 +1,1106 @@
+import inspect
+from typing import Any, Callable, Dict, List, Optional, Tuple, Union
+
+import numpy as np
+import PIL.Image
+import torch
+from diffusers import (AutoencoderKL, ControlNetModel, DiffusionPipeline,
+                       UNet2DConditionModel)
+from diffusers.loaders import LoraLoaderMixin
+from diffusers.pipelines.stable_diffusion import (
+    StableDiffusionPipelineOutput, StableDiffusionSafetyChecker)
+from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion_controlnet import \
+    MultiControlNetModel
+from diffusers.schedulers import KarrasDiffusionSchedulers
+from diffusers.utils import (PIL_INTERPOLATION, is_accelerate_available,
+                             is_accelerate_version, randn_tensor,
+                             replace_example_docstring)
+from transformers import CLIPImageProcessor, CLIPTextModel, CLIPTokenizer
+
+EXAMPLE_DOC_STRING = """
+    Examples:
+        ```py
+        >>> import numpy as np
+        >>> import torch
+        >>> from PIL import Image
+        >>> from diffusers import ControlNetModel, UniPCMultistepScheduler
+        >>> from diffusers.utils import load_image
+
+        >>> input_image = load_image("https://hf.co/datasets/huggingface/documentation-images/resolve/main/diffusers/input_image_vermeer.png")
+
+        >>> controlnet = ControlNetModel.from_pretrained("lllyasviel/sd-controlnet-canny", torch_dtype=torch.float16)
+
+        >>> pipe_controlnet = StableDiffusionControlNetImg2ImgPipeline.from_pretrained(
+                "runwayml/stable-diffusion-v1-5",
+                controlnet=controlnet,
+                safety_checker=None,
+                torch_dtype=torch.float16
+                )
+
+        >>> pipe_controlnet.scheduler = UniPCMultistepScheduler.from_config(pipe_controlnet.scheduler.config)
+        >>> pipe_controlnet.enable_xformers_memory_efficient_attention()
+        >>> pipe_controlnet.enable_model_cpu_offload()
+
+        # using image with edges for our canny controlnet
+        >>> control_image = load_image(
+            "https://hf.co/datasets/huggingface/documentation-images/resolve/main/diffusers/vermeer_canny_edged.png")
+
+
+        >>> result_img = pipe_controlnet(controlnet_conditioning_image=control_image,
+                        image=input_image,
+                        prompt="an android robot, cyberpank, digitl art masterpiece",
+                        num_inference_steps=20).images[0]
+
+        >>> result_img.show()
+        ```
+"""
+
+
+def prepare_image(image):
+    if isinstance(image, torch.Tensor):
+        # Batch single image
+        if image.ndim == 3:
+            image = image.unsqueeze(0)
+
+        image = image.to(dtype=torch.float32)
+    else:
+        # preprocess image
+        if isinstance(image, (PIL.Image.Image, np.ndarray)):
+            image = [image]
+
+        if isinstance(image, list) and isinstance(image[0], PIL.Image.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
+
+    return image
+
+
+def prepare_controlnet_conditioning_image(
+    controlnet_conditioning_image,
+    width,
+    height,
+    batch_size,
+    num_images_per_prompt,
+    device,
+    dtype,
+    do_classifier_free_guidance,
+):
+    if not isinstance(controlnet_conditioning_image, torch.Tensor):
+        if isinstance(controlnet_conditioning_image, PIL.Image.Image):
+            controlnet_conditioning_image = [controlnet_conditioning_image]
+
+        if isinstance(controlnet_conditioning_image[0], PIL.Image.Image):
+            controlnet_conditioning_image = [
+                np.array(
+                    i.resize((width, height), resample=PIL_INTERPOLATION["lanczos"])
+                )[None, :]
+                for i in controlnet_conditioning_image
+            ]
+            controlnet_conditioning_image = np.concatenate(
+                controlnet_conditioning_image, axis=0
+            )
+            controlnet_conditioning_image = (
+                np.array(controlnet_conditioning_image).astype(np.float32) / 255.0
+            )
+            controlnet_conditioning_image = controlnet_conditioning_image.transpose(
+                0, 3, 1, 2
+            )
+            controlnet_conditioning_image = torch.from_numpy(
+                controlnet_conditioning_image
+            )
+        elif isinstance(controlnet_conditioning_image[0], torch.Tensor):
+            controlnet_conditioning_image = torch.cat(
+                controlnet_conditioning_image, dim=0
+            )
+
+    image_batch_size = controlnet_conditioning_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
+
+    controlnet_conditioning_image = controlnet_conditioning_image.repeat_interleave(
+        repeat_by, dim=0
+    )
+
+    controlnet_conditioning_image = controlnet_conditioning_image.to(
+        device=device, dtype=dtype
+    )
+
+    if do_classifier_free_guidance:
+        controlnet_conditioning_image = torch.cat([controlnet_conditioning_image] * 2)
+
+    return controlnet_conditioning_image
+
+
+class StableDiffusionControlNetImg2ImgPipeline(DiffusionPipeline, LoraLoaderMixin):
+    """
+    Inspired by: https://github.com/haofanwang/ControlNet-for-Diffusers/
+    """
+
+    _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:
+            print(
+                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.register_to_config(requires_safety_checker=requires_safety_checker)
+
+    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()
+
+    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()
+
+    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
+    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
+
+    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:
+            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]
+                )
+                print(
+                    "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 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
+
+            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
+
+    def run_safety_checker(self, image, device, dtype):
+        if self.safety_checker is not None:
+            safety_checker_input = self.feature_extractor(
+                self.numpy_to_pil(image), return_tensors="pt"
+            ).to(device)
+            image, has_nsfw_concept = self.safety_checker(
+                images=image, clip_input=safety_checker_input.pixel_values.to(dtype)
+            )
+        else:
+            has_nsfw_concept = None
+        return image, has_nsfw_concept
+
+    def decode_latents(self, latents):
+        latents = 1 / self.vae.config.scaling_factor * latents
+        image = self.vae.decode(latents).sample
+        image = (image / 2 + 0.5).clamp(0, 1)
+        # we always cast to float32 as this does not cause significant overhead and is compatible with bfloat16
+        image = image.cpu().permute(0, 2, 3, 1).float().numpy()
+        return image
+
+    def prepare_extra_step_kwargs(self, generator, eta):
+        # prepare extra kwargs for the scheduler step, since not all schedulers have the same signature
+        # eta (η) is only used with the DDIMScheduler, it will be ignored for other schedulers.
+        # eta corresponds to η in DDIM paper: https://arxiv.org/abs/2010.02502
+        # and should be between [0, 1]
+
+        accepts_eta = "eta" in set(
+            inspect.signature(self.scheduler.step).parameters.keys()
+        )
+        extra_step_kwargs = {}
+        if accepts_eta:
+            extra_step_kwargs["eta"] = eta
+
+        # check if the scheduler accepts generator
+        accepts_generator = "generator" in set(
+            inspect.signature(self.scheduler.step).parameters.keys()
+        )
+        if accepts_generator:
+            extra_step_kwargs["generator"] = generator
+        return extra_step_kwargs
+
+    def check_controlnet_conditioning_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)
+        else:
+            raise ValueError("controlnet condition image is not valid")
+
+        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]
+        else:
+            raise ValueError("prompt or prompt_embeds are not valid")
+
+        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 check_inputs(
+        self,
+        prompt,
+        image,
+        controlnet_conditioning_image,
+        height,
+        width,
+        callback_steps,
+        negative_prompt=None,
+        prompt_embeds=None,
+        negative_prompt_embeds=None,
+        strength=None,
+        controlnet_guidance_start=None,
+        controlnet_guidance_end=None,
+        controlnet_conditioning_scale=None,
+    ):
+        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}."
+                )
+
+        # check controlnet condition image
+
+        if isinstance(self.controlnet, ControlNetModel):
+            self.check_controlnet_conditioning_image(
+                controlnet_conditioning_image, prompt, prompt_embeds
+            )
+        elif isinstance(self.controlnet, MultiControlNetModel):
+            if not isinstance(controlnet_conditioning_image, list):
+                raise TypeError("For multiple controlnets: `image` must be type `list`")
+
+            if len(controlnet_conditioning_image) != len(self.controlnet.nets):
+                raise ValueError(
+                    "For multiple controlnets: `image` must have the same length as the number of controlnets."
+                )
+
+            for image_ in controlnet_conditioning_image:
+                self.check_controlnet_conditioning_image(image_, prompt, prompt_embeds)
+        else:
+            assert False
+
+        # Check `controlnet_conditioning_scale`
+
+        if isinstance(self.controlnet, 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):
+            if 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
+
+        if isinstance(image, torch.Tensor):
+            if image.ndim != 3 and image.ndim != 4:
+                raise ValueError("`image` must have 3 or 4 dimensions")
+
+            if image.ndim == 3:
+                image_batch_size = 1
+                image_channels, image_height, image_width = image.shape
+            elif image.ndim == 4:
+                (
+                    image_batch_size,
+                    image_channels,
+                    image_height,
+                    image_width,
+                ) = image.shape
+            else:
+                assert False
+
+            if image_channels != 3:
+                raise ValueError("`image` must have 3 channels")
+
+            if image.min() < -1 or image.max() > 1:
+                raise ValueError("`image` should be in range [-1, 1]")
+
+        if self.vae.config.latent_channels != self.unet.config.in_channels:
+            raise ValueError(
+                f"The config of `pipeline.unet` expects {self.unet.config.in_channels} but received"
+                f" latent channels: {self.vae.config.latent_channels},"
+                f" Please verify the config of `pipeline.unet` and the `pipeline.vae`"
+            )
+
+        if strength < 0 or strength > 1:
+            raise ValueError(
+                f"The value of `strength` should in [0.0, 1.0] but is {strength}"
+            )
+
+        if controlnet_guidance_start < 0 or controlnet_guidance_start > 1:
+            raise ValueError(
+                f"The value of `controlnet_guidance_start` should in [0.0, 1.0] but is {controlnet_guidance_start}"
+            )
+
+        if controlnet_guidance_end < 0 or controlnet_guidance_end > 1:
+            raise ValueError(
+                f"The value of `controlnet_guidance_end` should in [0.0, 1.0] but is {controlnet_guidance_end}"
+            )
+
+        if controlnet_guidance_start > controlnet_guidance_end:
+            raise ValueError(
+                "The value of `controlnet_guidance_start` should be less than `controlnet_guidance_end`, but got"
+                f" `controlnet_guidance_start` {controlnet_guidance_start} >= `controlnet_guidance_end` {controlnet_guidance_end}"
+            )
+
+    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:]
+
+        return timesteps, num_inference_steps - t_start
+
+    def prepare_latents(
+        self,
+        image,
+        timestep,
+        batch_size,
+        num_images_per_prompt,
+        dtype,
+        device,
+        generator=None,
+    ):
+        if not isinstance(image, (torch.Tensor, PIL.Image.Image, list)):
+            raise ValueError(
+                f"`image` has to be of type `torch.Tensor`, `PIL.Image.Image` or list but is {type(image)}"
+            )
+
+        image = image.to(device=device, dtype=dtype)
+
+        batch_size = batch_size * num_images_per_prompt
+        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 isinstance(generator, list):
+            init_latents = [
+                self.vae.encode(image[i : i + 1]).latent_dist.sample(generator[i])
+                for i in range(batch_size)
+            ]
+            init_latents = torch.cat(init_latents, dim=0)
+        else:
+            init_latents = self.vae.encode(image).latent_dist.sample(generator)
+
+        init_latents = self.vae.config.scaling_factor * init_latents
+
+        if (
+            batch_size > init_latents.shape[0]
+            and batch_size % init_latents.shape[0] == 0
+        ):
+            raise ValueError(
+                f"Cannot duplicate `image` of batch size {init_latents.shape[0]} to {batch_size} text prompts."
+            )
+        else:
+            init_latents = torch.cat([init_latents], dim=0)
+
+        shape = init_latents.shape
+        noise = randn_tensor(shape, generator=generator, device=device, dtype=dtype)
+
+        # get latents
+        init_latents = self.scheduler.add_noise(init_latents, noise, timestep)
+        latents = init_latents
+
+        return latents
+
+    def _default_height_width(self, height, width, image):
+        if 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[3]
+
+            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[2]
+
+            width = (width // 8) * 8  # round down to nearest multiple of 8
+
+        return height, width
+
+    @torch.no_grad()
+    @replace_example_docstring(EXAMPLE_DOC_STRING)
+    def __call__(
+        self,
+        prompt: Union[str, List[str]] = None,
+        image: Union[torch.Tensor, PIL.Image.Image] = None,
+        controlnet_conditioning_image: Union[
+            torch.FloatTensor,
+            PIL.Image.Image,
+            List[torch.FloatTensor],
+            List[PIL.Image.Image],
+        ] = None,
+        strength: float = 0.8,
+        height: Optional[int] = None,
+        width: Optional[int] = None,
+        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]] = 1.0,
+        controlnet_guidance_start: float = 0.0,
+        controlnet_guidance_end: float = 1.0,
+    ):
+        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.Tensor` or `PIL.Image.Image`):
+                `Image`, or tensor representing an image batch which will be inpainted, *i.e.* parts of the image will
+                be masked out with `mask_image` and repainted according to `prompt`.
+            controlnet_conditioning_image (`torch.FloatTensor`, `PIL.Image.Image`, `List[torch.FloatTensor]` or `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 control image is automatically resized to fit the output image.
+            strength (`float`, *optional*):
+                Conceptually, indicates how much to transform the reference `image`. Must be between 0 and 1. `image`
+                will be used as a starting point, adding more noise to it the larger the `strength`. The number of
+                denoising steps depends on the amount of noise initially added. When `strength` is 1, added noise will
+                be maximum and the denoising process will run for the full number of iterations specified in
+                `num_inference_steps`. A value of 1, therefore, essentially ignores `image`.
+            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`, *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.
+            controlnet_guidance_start ('float', *optional*, defaults to 0.0):
+                The percentage of total steps the controlnet starts applying. Must be between 0 and 1.
+            controlnet_guidance_end ('float', *optional*, defaults to 1.0):
+                The percentage of total steps the controlnet ends applying. Must be between 0 and 1. Must be greater
+                than `controlnet_guidance_start`.
+
+        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, controlnet_conditioning_image
+        )
+
+        # 1. Check inputs. Raise error if not correct
+        self.check_inputs(
+            prompt,
+            image,
+            controlnet_conditioning_image,
+            height,
+            width,
+            callback_steps,
+            negative_prompt,
+            prompt_embeds,
+            negative_prompt_embeds,
+            strength,
+            controlnet_guidance_start,
+            controlnet_guidance_end,
+            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
+
+        if isinstance(self.controlnet, MultiControlNetModel) and isinstance(
+            controlnet_conditioning_scale, float
+        ):
+            controlnet_conditioning_scale = [controlnet_conditioning_scale] * len(
+                self.controlnet.nets
+            )
+
+        # 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)
+
+        # condition image(s)
+        if isinstance(self.controlnet, ControlNetModel):
+            controlnet_conditioning_image = prepare_controlnet_conditioning_image(
+                controlnet_conditioning_image=controlnet_conditioning_image,
+                width=width,
+                height=height,
+                batch_size=batch_size * num_images_per_prompt,
+                num_images_per_prompt=num_images_per_prompt,
+                device=device,
+                dtype=self.controlnet.dtype,
+                do_classifier_free_guidance=do_classifier_free_guidance,
+            )
+        elif isinstance(self.controlnet, MultiControlNetModel):
+            controlnet_conditioning_images = []
+
+            for image_ in controlnet_conditioning_image:
+                image_ = prepare_controlnet_conditioning_image(
+                    controlnet_conditioning_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=self.controlnet.dtype,
+                    do_classifier_free_guidance=do_classifier_free_guidance,
+                )
+
+                controlnet_conditioning_images.append(image_)
+
+            controlnet_conditioning_image = controlnet_conditioning_images
+        else:
+            assert False
+
+        # 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)
+
+        # 6. Prepare latent variables
+        latents = self.prepare_latents(
+            image,
+            latent_timestep,
+            batch_size,
+            num_images_per_prompt,
+            prompt_embeds.dtype,
+            device,
+            generator,
+        )
+
+        # 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
+                )
+
+                # compute the percentage of total steps we are at
+                current_sampling_percent = i / len(timesteps)
+
+                if (
+                    current_sampling_percent < controlnet_guidance_start
+                    or current_sampling_percent > controlnet_guidance_end
+                ):
+                    # do not apply the controlnet
+                    down_block_res_samples = None
+                    mid_block_res_sample = None
+                else:
+                    # apply the controlnet
+                    down_block_res_samples, mid_block_res_sample = self.controlnet(
+                        latent_model_input,
+                        t,
+                        encoder_hidden_states=prompt_embeds,
+                        controlnet_cond=controlnet_conditioning_image,
+                        conditioning_scale=controlnet_conditioning_scale,
+                        return_dict=False,
+                    )
+
+                # 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,
+                ).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
+                ).prev_sample
+
+                # 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 output_type == "latent":
+            image = latents
+            has_nsfw_concept = None
+        elif output_type == "pil":
+            # 8. Post-processing
+            image = self.decode_latents(latents)
+
+            # 9. Run safety checker
+            image, has_nsfw_concept = self.run_safety_checker(
+                image, device, prompt_embeds.dtype
+            )
+
+            # 10. Convert to PIL
+            image = self.numpy_to_pil(image)
+        else:
+            # 8. Post-processing
+            image = self.decode_latents(latents)
+
+            # 9. Run safety checker
+            image, has_nsfw_concept = self.run_safety_checker(
+                image, device, prompt_embeds.dtype
+            )
+
+        # 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
+        )

internals/pipelines/twoStepPipeline.py

@@ -250,3 +250,292 @@ class two_step_pipeline(StableDiffusionPipeline):
         return StableDiffusionPipelineOutput(
             images=image, nsfw_content_detected=has_nsfw_concept
         )
+
+    @torch.no_grad()
+    def multi_character_diffusion(
+        self,
+        prompt: Union[str, List[str]],
+        pos: List[str],
+        mix_val: Union[float, List[float]] = 0.5,
+        height: Optional[int] = None,
+        width: Optional[int] = None,
+        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[torch.Generator] = None,
+        latents: Optional[torch.FloatTensor] = None,
+        output_type: Optional[str] = "pil",
+        return_dict: bool = True,
+        callback: Optional[Callable[[int, int, torch.FloatTensor], None]] = None,
+        callback_steps: Optional[int] = 1,
+    ):
+        r"""
+        Function invoked when calling the pipeline for generation.
+
+        Args:
+            prompt (`str` or `List[str]`):
+                The prompt or prompts to guide the image generation.
+            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. 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`, *optional*):
+                A [torch generator](https://pytorch.org/docs/stable/generated/torch.Generator.html) to make generation
+                deterministic.
+            latents (`torch.FloatTensor`, *optional*):
+                Pre-generated noisy latents, sampled from a Gaussian distribution, to be used as inputs for image
+                generation. Can be used to tweak the same generation with different prompts. If not provided, a latents
+                tensor will ge generated by sampling using the supplied random `generator`.
+            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.
+
+        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`.
+        """
+        # 生成する画像サイズは8で割り切れなければならない
+        height = height - height % 8
+        width = width - width % 8
+
+        # 0. Default height and width to unet
+        height = height or self.unet.config.sample_size * self.vae_scale_factor
+        width = width or self.unet.config.sample_size * self.vae_scale_factor
+
+        # 1. Check inputs. Raise error if not correct
+        self.check_inputs(prompt[0], height, width, callback_steps)
+
+        # 2. Define call parameters
+        batch_size = 1 if isinstance(prompt[0], str) else len(prompt[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_embeddings = []
+        for i in range(len(prompt)):
+            one_text_embeddings = self._encode_prompt(
+                prompt[i],
+                device,
+                num_images_per_prompt,
+                do_classifier_free_guidance,
+                negative_prompt[i],
+            )
+            text_embeddings.append(one_text_embeddings)
+
+        # 4. Prepare timesteps
+        self.scheduler.set_timesteps(num_inference_steps, device=device)
+        timesteps = self.scheduler.timesteps
+
+        # 5. Prepare latent variables
+        num_channels_latents = self.unet.in_channels
+        latents = self.prepare_latents(
+            batch_size * num_images_per_prompt,
+            num_channels_latents,
+            height,
+            width,
+            text_embeddings[0].dtype,
+            device,
+            generator,
+            latents,
+        )
+
+        # 6. Prepare extra step kwargs. TODO: Logic should ideally just be moved out of the pipeline
+        extra_step_kwargs = self.prepare_extra_step_kwargs(generator, eta)
+
+        # 7. Denoising loop
+        # num_warmup_steps = len(timesteps) - num_inference_steps# * self.scheduler.order
+        for i, t in enumerate(self.progress_bar(timesteps)):
+            # expand the latents if we are doing classifier free guidance
+            latent_model_input = (
+                torch.cat([latents] * 2) if do_classifier_free_guidance else latents
+            )
+            latent_model_input = self.scheduler.scale_model_input(latent_model_input, t)
+
+            # predict the noise residual
+            noise_preds = []
+            for i in range(len(prompt)):
+                noise_pred = self.unet(
+                    latent_model_input, t, encoder_hidden_states=text_embeddings[i]
+                ).sample
+                noise_preds.append(noise_pred)
+            # perform guidance
+            if do_classifier_free_guidance:
+                noise_pred_unconds = []
+                noise_pred_texts = []
+                for i in range(len(prompt)):
+                    noise_pred_uncond, noise_pred_text = noise_preds[i].chunk(2)
+                    noise_pred_unconds.append(noise_pred_uncond)
+                    noise_pred_texts.append(noise_pred_text)
+                # TODO：posに基づいてフィルターを作る
+                mask_list = []
+                for i in range(len(prompt)):
+                    pos_base = pos[i].split("-")
+                    pos_dev = pos_base[0].split(":")  # 1:2
+                    pos_pos = pos_base[1].split(":")  # 0:0
+                    one_filter = None
+                    zero_f = False
+                    for y in range(int(pos_dev[0])):
+                        one_line = None
+                        zero = False
+                        for x in range(int(pos_dev[1])):
+                            if y == int(pos_pos[0]) and x == int(pos_pos[1]):
+                                # print("same", zero, (height//8) // int(pos_dev[0]), (width//8) // int(pos_dev[1]))
+                                if zero:
+                                    one_block = (
+                                        torch.ones(
+                                            batch_size,
+                                            4,
+                                            (height // 8) // int(pos_dev[0]),
+                                            (width // 8) // int(pos_dev[1]),
+                                        )
+                                        .to(device)
+                                        .to(torch.float16)
+                                        * mix_val[i]
+                                    )
+                                    one_line = torch.cat((one_line, one_block), 3)
+                                else:
+                                    zero = True
+                                    one_block = (
+                                        torch.ones(
+                                            batch_size,
+                                            4,
+                                            (height // 8) // int(pos_dev[0]),
+                                            (width // 8) // int(pos_dev[1]),
+                                        )
+                                        .to(device)
+                                        .to(torch.float16)
+                                        * mix_val[i]
+                                    )
+                                    one_line = one_block
+                            else:
+                                # print("else", zero, (height//8) // int(pos_dev[0]), (width//8) // int(pos_dev[1]))
+                                if zero:
+                                    one_block = (
+                                        torch.zeros(
+                                            batch_size,
+                                            4,
+                                            (height // 8) // int(pos_dev[0]),
+                                            (width // 8) // int(pos_dev[1]),
+                                        )
+                                        .to(device)
+                                        .to(torch.float16)
+                                    )
+                                    one_line = torch.cat((one_line, one_block), 3)
+                                else:
+                                    zero = True
+                                    one_block = (
+                                        torch.zeros(
+                                            batch_size,
+                                            4,
+                                            (height // 8) // int(pos_dev[0]),
+                                            (width // 8) // int(pos_dev[1]),
+                                        )
+                                        .to(device)
+                                        .to(torch.float16)
+                                    )
+                                    one_line = one_block
+                        one_block = (
+                            torch.zeros(
+                                batch_size,
+                                4,
+                                (height // 8) // int(pos_dev[0]),
+                                (width // 8) - one_line.size()[3],
+                            )
+                            .to(device)
+                            .to(torch.float16)
+                        )
+                        one_line = torch.cat((one_line, one_block), 3)
+                        if zero_f:
+                            one_filter = torch.cat((one_filter, one_line), 2)
+                        else:
+                            zero_f = True
+                            one_filter = one_line
+                    mask_list.append(one_filter)
+                for i in range(len(mask_list)):
+                    import torchvision
+
+                    torchvision.transforms.functional.to_pil_image(
+                        mask_list[i][0] * 256
+                    ).save(str(i) + ".png")
+
+                result = None
+                noise_preds = []
+                for i in range(len(prompt)):
+                    noise_pred = noise_pred_unconds[i] + guidance_scale * (
+                        noise_pred_texts[i] - noise_pred_unconds[i]
+                    )
+                    noise_preds.append(noise_pred)
+                result = noise_preds[0] * mask_list[0]
+                for i in range(1, len(prompt)):
+                    result += noise_preds[i] * mask_list[i]
+
+                # 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(
+                    result, t, latents, **extra_step_kwargs
+                ).prev_sample
+
+                # call the callback, if provided
+                if callback is not None and i % callback_steps == 0:
+                    callback(i, t, latents)
+
+        # 8. Post-processing
+        image = self.decode_latents(latents)
+
+        # 9. Run safety checker
+        image, has_nsfw_concept = self.run_safety_checker(
+            image, device, text_embeddings[0].dtype
+        )
+
+        # 10. Convert to PIL
+        if output_type == "pil":
+            image = self.numpy_to_pil(image)
+            output = []
+            import torchvision
+
+            for i in mask_list:
+                output.append(
+                    torchvision.transforms.functional.to_pil_image(i[0] * 256)
+                )
+
+        if not return_dict:
+            return (image, has_nsfw_concept)
+
+        return StableDiffusionPipelineOutput(
+            images=image, nsfw_content_detected=has_nsfw_concept
+        )

internals/util/config.py

@@ -8,6 +8,8 @@ nsfw_access = False
 access_token = ""
 root_dir = ""
 
+num_return_sequences = 4  # the number of results to generate
+
 
 def set_root_dir(main_file: str):
     global root_dir

internals/util/lora_style.py

@@ -5,6 +5,7 @@ from typing import Any, Dict, List, Union
 
 import boto3
 import torch
+from diffusers.models.attention_processor import AttnProcessor2_0
 from lora_diffusion import patch_pipe, tune_lora_scale
 from pydash import chain
 
@@ -32,7 +33,24 @@ class LoraStyle:
         def cleanup(self):
             tune_lora_scale(self.pipe.unet, 0.0)
             tune_lora_scale(self.pipe.text_encoder, 0.0)
-            pass
+
+    class LoraDiffuserPatcher:
+        def __init__(self, pipe, style: Dict[str, Any]):
+            self.__style = style
+            self.pipe = pipe
+
+        @torch.inference_mode()
+        def patch(self):
+            path = self.__style["path"]
+            self.pipe.load_lora_weights(
+                os.path.dirname(path), weight_name=os.path.basename(path)
+            )
+
+        def kwargs(self):
+            return {}
+
+        def cleanup(self):
+            LoraStyle.unload_lora_weights(self.pipe)
 
     class EmptyLoraPatcher:
         def __init__(self, pipe):
@@ -41,7 +59,6 @@ class LoraStyle:
         def patch(self):
             "Patch will act as cleanup, to tune down any corrupted lora"
             self.cleanup()
-            pass
 
         def kwargs(self):
             return {}
@@ -49,7 +66,7 @@ class LoraStyle:
         def cleanup(self):
             tune_lora_scale(self.pipe.unet, 0.0)
             tune_lora_scale(self.pipe.text_encoder, 0.0)
-            pass
+            LoraStyle.unload_lora_weights(self.pipe)
 
     def load(self, model_dir: str):
         self.model = model_dir
@@ -70,9 +87,13 @@ class LoraStyle:
             return f"{', '.join(style['text'])}, {prompt}"
         return prompt
 
-    def get_patcher(self, pipe, key: str) -> Union[LoraPatcher, EmptyLoraPatcher]:
+    def get_patcher(
+        self, pipe, key: str
+    ) -> Union[LoraPatcher, LoraDiffuserPatcher, EmptyLoraPatcher]:
         if key in self.__styles:
             style = self.__styles[key]
+            if style["type"] == "diffuser":
+                return self.LoraDiffuserPatcher(pipe, style)
             return self.LoraPatcher(pipe, style)
         return self.EmptyLoraPatcher(pipe)
 
@@ -152,3 +173,8 @@ class LoraStyle:
                     + " not found at path: "
                     + self.__styles[item]["path"]
                 )
+
+    @staticmethod
+    def unload_lora_weights(pipe):
+        pipe.unet.set_attn_processor(AttnProcessor2_0())  # for pytorch 2.0
+        pipe._remove_text_encoder_monkey_patch()

requirements.txt

@@ -8,7 +8,6 @@ redis==4.3.4
 requests==2.28.1
 transformers
 rembg==2.0.30
-accelerate==0.17.0
 gfpgan==1.3.8
 rembg==2.0.30
 controlnet-aux==0.0.5
@@ -20,6 +19,7 @@ albumentations==0.5.2
 kornia==0.5.0
 pytorch-lightning==1.2.9
 pydash
+accelerate
 pandas
 xformers
 torchvision