| from PIL import Image |
| import torch |
| import requests |
| from io import BytesIO |
| import numpy as np |
| import fal_client |
|
|
| class FalDiffusion: |
| @classmethod |
| def INPUT_TYPES(s): |
| return { |
| "required": { |
| "prompt": ("STRING", {"multiline": False}), |
| "steps": ("INT",{"default": 2, "min": 1, "max": 8, "step": 1}), |
| } |
| } |
|
|
| RETURN_TYPES = ("IMAGE",) |
| CATEGORY = "external_tooling" |
| FUNCTION = "load" |
|
|
| def load(self, prompt: str, steps: int): |
| |
| handler = fal_client.submit( |
| "fal-ai/flux/schnell", |
| arguments={ |
| "prompt": f"{prompt}", |
| "image_size": "square_hd", |
| "num_inference_steps": steps, |
| }, |
| ) |
| result = handler.get() |
| images = [] |
| for image in result['images']: |
| url = image['url'] |
| response = requests.get(url) |
| img = Image.open(BytesIO(response.content)) |
| img = np.array(img).astype(np.float32) / 255.0 |
| img = torch.from_numpy(img)[None,] |
| images.append(img) |
| return (torch.cat(images, dim=0),) |
|
|
|
|
| class FalDifferentialDiffusion: |
| @classmethod |
| def INPUT_TYPES(s): |
| return { |
| "required": { |
| "foreground_image": ("IMAGE", ), |
| "depth_image": ("IMAGE", ), |
| "foreground_prompt": ("STRING", {"multiline": False}), |
| "background_prompt": ("STRING", {"multiline": False}), |
| "strength": ("FLOAT",{"default": 1, "min": 0.01, "max": 3, "step": 0.01}), |
| "steps": ("INT",{"default": 14, "min": 1, "max": 32, "step": 1}), |
| } |
| } |
|
|
| RETURN_TYPES = ("IMAGE",) |
| CATEGORY = "external_tooling" |
| FUNCTION = "load" |
|
|
| def load(self, foreground_image: torch.Tensor, depth_image: torch.Tensor, foreground_prompt: str, background_prompt: str, strength: float, steps: int): |
| |
| foreground_image_array = foreground_image.squeeze(0).cpu().numpy() * 255.0 |
| foreground_image_pil = Image.fromarray(np.clip(foreground_image_array, 0, 255).astype(np.uint8)) |
| foreground_output = BytesIO() |
| foreground_image_pil.save(foreground_output, format='PNG') |
| foreground_url = fal_client.upload(foreground_output.getvalue(), "image/png") |
| |
| depth_image_array = depth_image.squeeze(0).cpu().numpy() * 255.0 |
| depth_image_pil = Image.fromarray(np.clip(depth_image_array, 0, 255).astype(np.uint8)) |
| depth_output = BytesIO() |
| depth_image_pil.save(depth_output, format='PNG') |
| depth_url = fal_client.upload(depth_output.getvalue(), "image/png") |
| |
| handler = fal_client.submit( |
| "fal-ai/flux-differential-diffusion", |
| arguments={ |
| "prompt": f"{foreground_prompt}, {background_prompt}, 8k, unreal engine 5, hightly detailed, intricate detailed.", |
| "image_url": foreground_url, |
| "change_map_image_url": depth_url, |
| "strength": strength, |
| "num_inference_steps": steps, |
| }, |
| ) |
| result = handler.get() |
| images = [] |
| for image in result['images']: |
| url = image['url'] |
| response = requests.get(url) |
| img = Image.open(BytesIO(response.content)) |
| img = np.array(img).astype(np.float32) / 255.0 |
| img = torch.from_numpy(img)[None,] |
| images.append(img) |
| return (torch.cat(images, dim=0),) |
