| import numpy as np | |
| import torch | |
| from transformers.tools.base import Tool, get_default_device | |
| from transformers.utils import ( | |
| is_accelerate_available, | |
| is_diffusers_available, | |
| is_opencv_available, | |
| is_vision_available, | |
| ) | |
| if is_vision_available(): | |
| from PIL import Image | |
| if is_diffusers_available(): | |
| from diffusers import ControlNetModel, StableDiffusionControlNetPipeline, UniPCMultistepScheduler | |
| if is_opencv_available(): | |
| import cv2 | |
| IMAGE_TRANSFORMATION_DESCRIPTION = ( | |
| "This is a tool that transforms an image according to a prompt. It takes two inputs: `image`, which should be " | |
| "the image to transform, and `prompt`, which should be the prompt to use to change it. It returns the " | |
| "modified image." | |
| ) | |
| class ImageTransformationTool(Tool): | |
| default_stable_diffusion_checkpoint = "runwayml/stable-diffusion-v1-5" | |
| default_controlnet_checkpoint = "lllyasviel/control_v11p_sd15_canny" | |
| description = IMAGE_TRANSFORMATION_DESCRIPTION | |
| inputs = ['image', 'text'] | |
| outputs = ['image'] | |
| def __init__(self, device=None, controlnet=None, stable_diffusion=None, **hub_kwargs) -> None: | |
| if not is_accelerate_available(): | |
| raise ImportError("Accelerate should be installed in order to use tools.") | |
| if not is_diffusers_available(): | |
| raise ImportError("Diffusers should be installed in order to use the StableDiffusionTool.") | |
| if not is_vision_available(): | |
| raise ImportError("Pillow should be installed in order to use the StableDiffusionTool.") | |
| if not is_opencv_available(): | |
| raise ImportError("opencv should be installed in order to use the StableDiffusionTool.") | |
| super().__init__() | |
| if controlnet is None: | |
| controlnet = self.default_controlnet_checkpoint | |
| self.controlnet_checkpoint = controlnet | |
| if stable_diffusion is None: | |
| stable_diffusion = self.default_stable_diffusion_checkpoint | |
| self.stable_diffusion_checkpoint = stable_diffusion | |
| self.device = device | |
| self.hub_kwargs = hub_kwargs | |
| def setup(self): | |
| if self.device is None: | |
| self.device = get_default_device() | |
| self.controlnet = ControlNetModel.from_pretrained(self.controlnet_checkpoint) | |
| self.pipeline = StableDiffusionControlNetPipeline.from_pretrained( | |
| self.stable_diffusion_checkpoint, controlnet=self.controlnet | |
| ) | |
| self.pipeline.scheduler = UniPCMultistepScheduler.from_config(self.pipeline.scheduler.config) | |
| self.pipeline.to(device=self.device) | |
| if self.device.type == "cuda": | |
| self.pipeline.to(torch_dtype=torch.float16) | |
| self.is_initialized = True | |
| def __call__(self, image, prompt, negative_prompt="low quality, bad quality, deformed, low resolution", added_prompt=" , highest quality, highly realistic, very high resolution"): | |
| if not self.is_initialized: | |
| self.setup() | |
| initial_prompt = "super-hero character, best quality, extremely detailed" | |
| prompt = initial_prompt + prompt | |
| low_threshold = 100 | |
| high_threshold = 200 | |
| image = np.array(image) | |
| image = cv2.Canny(image, low_threshold, high_threshold) | |
| image = image[:, :, None] | |
| image = np.concatenate([image, image, image], axis=2) | |
| canny_image = Image.fromarray(image) | |
| return self.pipeline( | |
| prompt + added_prompt, | |
| canny_image, | |
| negative_prompt=negative_prompt, | |
| num_inference_steps=25, | |
| ).images[0] | |