Spaces:
Running
on
A10G
Running
on
A10G
File size: 3,671 Bytes
45efc4b e11e948 45efc4b b6d3fe5 45efc4b b6d3fe5 45efc4b b6d3fe5 45efc4b b6d3fe5 45efc4b b6d3fe5 45efc4b |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 |
import torch
from src.config import RunConfig
import PIL
from src.euler_scheduler import MyEulerAncestralDiscreteScheduler
from diffusers.pipelines.auto_pipeline import AutoPipelineForImage2Image
from src.sdxl_inversion_pipeline import SDXLDDIMPipeline
from diffusers.utils.torch_utils import randn_tensor
def inversion_callback(pipe, step, timestep, callback_kwargs):
return callback_kwargs
def inference_callback(pipe, step, timestep, callback_kwargs):
return callback_kwargs
def center_crop(im):
width, height = im.size # Get dimensions
min_dim = min(width, height)
left = (width - min_dim) / 2
top = (height - min_dim) / 2
right = (width + min_dim) / 2
bottom = (height + min_dim) / 2
# Crop the center of the image
im = im.crop((left, top, right, bottom))
return im
def load_im_into_format_from_path(im_path):
if isinstance(im_path, str):
return center_crop(PIL.Image.open(im_path)).resize((512, 512))
else:
return center_crop(im_path).resize((512, 512))
class ImageEditorDemo:
def __init__(self, pipe_inversion, pipe_inference, input_image, description_prompt, cfg, device):
self.pipe_inversion = pipe_inversion
self.pipe_inference = pipe_inference
self.original_image = load_im_into_format_from_path(input_image).convert("RGB")
self.load_image = True
g_cpu = torch.Generator().manual_seed(7865)
img_size = (512,512)
VQAE_SCALE = 8
latents_size = (1, 4, img_size[0] // VQAE_SCALE, img_size[1] // VQAE_SCALE)
noise = [randn_tensor(latents_size, dtype=torch.float16, device=torch.device(device), generator=g_cpu) for i
in range(cfg.num_inversion_steps)]
pipe_inversion.scheduler.set_noise_list(noise)
pipe_inference.scheduler.set_noise_list(noise)
pipe_inversion.scheduler_inference.set_noise_list(noise)
pipe_inversion.set_progress_bar_config(disable=True)
pipe_inference.set_progress_bar_config(disable=True)
self.cfg = cfg
self.pipe_inversion.cfg = cfg
self.pipe_inference.cfg = cfg
self.inv_hp = [2, 0.1, 0.2]
self.edit_cfg = cfg.edit_guidance_scale
self.pipe_inference.to(device)
self.pipe_inversion.to(device)
self.last_latent = self.invert(self.original_image, description_prompt)
self.original_latent = self.last_latent
def invert(self, init_image, base_prompt):
res = self.pipe_inversion(prompt=base_prompt,
num_inversion_steps=self.cfg.num_inversion_steps,
num_inference_steps=self.cfg.num_inference_steps,
image=init_image,
guidance_scale=self.cfg.inversion_guidance_scale,
callback_on_step_end=inversion_callback,
strength=self.cfg.inversion_max_step,
denoising_start=1.0 - self.cfg.inversion_max_step,
inv_hp=self.inv_hp)[0][0]
return res
def edit(self, target_prompt):
image = self.pipe_inference(prompt=target_prompt,
num_inference_steps=self.cfg.num_inference_steps,
negative_prompt="",
callback_on_step_end=inference_callback,
image=self.last_latent,
strength=self.cfg.inversion_max_step,
denoising_start=1.0 - self.cfg.inversion_max_step,
guidance_scale=self.edit_cfg).images[0]
return image
|