| import ipdb |
| import torch |
| from typing import Optional, Union, List, Tuple, Any |
|
|
| from torch import Tensor |
|
|
| from diffusers.pipelines import FluxPipeline |
| from PIL import Image, ImageFilter |
| import numpy as np |
| import cv2 |
|
|
| condition_dict = { |
| "depth": 0, |
| "canny": 1, |
| "subject": 4, |
| "coloring": 6, |
| "deblurring": 7, |
| "fill": 9, |
| } |
|
|
|
|
| class Condition(object): |
| def __init__( |
| self, |
| condition_type: str, |
| raw_img: Union[Image.Image, torch.Tensor] = None, |
| no_process: bool = False, |
| condition: Union[Image.Image, torch.Tensor] = None, |
| condition_ids = None, |
| mask=None, |
| ) -> None: |
| self.condition_type = condition_type |
| assert raw_img is not None or condition is not None |
| if raw_img is not None: |
| if no_process: |
| self.condition = raw_img.convert("RGB") |
| else: |
| self.condition = self.get_condition(condition_type, raw_img) |
| self.condition_ids = None |
| else: |
| self.condition = condition |
| self.condition_ids = condition_ids |
| |
| assert mask is None, "Mask not supported yet" |
|
|
| def get_condition( |
| self, condition_type: str, raw_img: Union[Image.Image, torch.Tensor] |
| ) -> Union[Image.Image, torch.Tensor]: |
| """ |
| Returns the condition image. |
| """ |
| if condition_type == "depth": |
| from transformers import pipeline |
|
|
| depth_pipe = pipeline( |
| task="depth-estimation", |
| model="ckpt/depth-anything-small-hf", |
| device="cuda", |
| ) |
| source_image = raw_img.convert("RGB") |
| condition_img = depth_pipe(source_image)["depth"].convert("RGB") |
| return condition_img |
| elif condition_type == "canny": |
| img = np.array(raw_img) |
| edges = cv2.Canny(img, 100, 200) |
| edges = Image.fromarray(edges).convert("RGB") |
| return edges |
| elif condition_type == "subject": |
| return raw_img |
| elif condition_type == "coloring": |
| return raw_img.convert("L").convert("RGB") |
| elif condition_type == "deblurring": |
| condition_image = ( |
| raw_img.convert("RGB") |
| .filter(ImageFilter.GaussianBlur(10)) |
| .convert("RGB") |
| ) |
| return condition_image |
| elif condition_type == "fill": |
| return raw_img.convert("RGB") |
| return self.condition |
|
|
| @property |
| def type_id(self) -> int: |
| """ |
| Returns the type id of the condition. |
| """ |
| return condition_dict[self.condition_type] |
|
|
| def _encode_image(self, pipe: FluxPipeline, cond_img: Image.Image) -> tuple[Any, Any]: |
| """ |
| Encodes an image condition into tokens using the pipeline. |
| """ |
| cond_img = pipe.image_processor.preprocess(cond_img) |
| cond_img = cond_img.to("cuda").to(pipe.dtype) |
| |
| cond_img = pipe.vae.encode(cond_img).latent_dist.sample() |
| cond_img = ( |
| cond_img - pipe.vae.config.shift_factor |
| ) * pipe.vae.config.scaling_factor |
| cond_tokens = pipe._pack_latents(cond_img, *cond_img.shape) |
| cond_ids = pipe._prepare_latent_image_ids( |
| cond_img.shape[0], |
| cond_img.shape[2]//2, |
| cond_img.shape[3]//2, |
| pipe.device, |
| pipe.dtype, |
| ) |
| if self.condition_type == "subject": |
| cond_ids[:, 2] += cond_img.shape[2] //2 |
| return cond_tokens, cond_ids |
|
|
| def encode(self, pipe: FluxPipeline) -> tuple[Any, Any, Tensor]: |
| """ |
| Encodes the condition into tokens, ids and type_id. |
| """ |
| if self.condition_ids is not None: |
| tokens, ids = self.condition, self.condition_ids |
| elif self.condition_ids is None and self.condition_type in [ |
| "depth", |
| "canny", |
| "subject", |
| "coloring", |
| "deblurring", |
| "fill", |
| ]: |
| tokens, ids = self._encode_image(pipe, self.condition) |
| else: |
| raise NotImplementedError( |
| f"There are two ways to use it: \n" |
| f"(1) Give the condition tensor to the 'self.condition' and the condition_ids to the 'self.condition_ids' manually.\n" |
| f"(2) Give the raw_image to the 'self.raw_img' and process the rest operations with a pipeline automatically.\n" |
| ) |
| type_id = torch.ones_like(ids[:, :1]) * self.type_id |
| return tokens, ids, type_id |
|
|
