Spaces:
Running
on
Zero
Running
on
Zero
File size: 7,030 Bytes
edb0494 6405936 edb0494 6405936 edb0494 a7d8817 d49f90c a7d8817 6405936 0e8df3d 6e4f1a9 6405936 49f2888 a7d8817 b230b71 a7d8817 80b786b a7d8817 6405936 a7d8817 6405936 a7d8817 6e4f1a9 0e8df3d 6e4f1a9 d49f90c 6e4f1a9 d49f90c 6e4f1a9 d49f90c 6e4f1a9 d49f90c 6e4f1a9 d49f90c 6e4f1a9 d49f90c 6e4f1a9 d49f90c 6e4f1a9 d49f90c 6e4f1a9 6405936 6e4f1a9 6405936 97567b1 6405936 97567b1 6405936 97567b1 6405936 b230b71 6405936 97567b1 6405936 |
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 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 |
import gradio as gr
import spaces
import torch
from diffusers import AutoencoderKL, TCDScheduler
from diffusers.models.model_loading_utils import load_state_dict
from gradio_imageslider import ImageSlider
from huggingface_hub import hf_hub_download
from controlnet_union import ControlNetModel_Union
from pipeline_fill_sd_xl import StableDiffusionXLFillPipeline
from PIL import Image, ImageDraw
import numpy as np
MODELS = {
"RealVisXL V5.0 Lightning": "SG161222/RealVisXL_V5.0_Lightning",
}
config_file = hf_hub_download(
"xinsir/controlnet-union-sdxl-1.0",
filename="config_promax.json",
)
config = ControlNetModel_Union.load_config(config_file)
controlnet_model = ControlNetModel_Union.from_config(config)
model_file = hf_hub_download(
"xinsir/controlnet-union-sdxl-1.0",
filename="diffusion_pytorch_model_promax.safetensors",
)
state_dict = load_state_dict(model_file)
model, _, _, _, _ = ControlNetModel_Union._load_pretrained_model(
controlnet_model, state_dict, model_file, "xinsir/controlnet-union-sdxl-1.0"
)
model.to(device="cuda", dtype=torch.float16)
vae = AutoencoderKL.from_pretrained(
"madebyollin/sdxl-vae-fp16-fix", torch_dtype=torch.float16
).to("cuda")
pipe = StableDiffusionXLFillPipeline.from_pretrained(
"SG161222/RealVisXL_V5.0_Lightning",
torch_dtype=torch.float16,
vae=vae,
controlnet=model,
variant="fp16",
).to("cuda")
pipe.scheduler = TCDScheduler.from_config(pipe.scheduler.config)
prompt = "high quality"
(
prompt_embeds,
negative_prompt_embeds,
pooled_prompt_embeds,
negative_pooled_prompt_embeds,
) = pipe.encode_prompt(prompt, "cuda", True)
"""
def fill_image(image, model_selection):
margin = 256
overlap = 24
# Open the original image
source = image # Changed from image["background"] to match new input format
# Calculate new output size
output_size = (source.width + 2*margin, source.height + 2*margin)
# Create a white background
background = Image.new('RGB', output_size, (255, 255, 255))
# Calculate position to paste the original image
position = (margin, margin)
# Paste the original image onto the white background
background.paste(source, position)
# Create the mask
mask = Image.new('L', output_size, 255) # Start with all white
mask_draw = ImageDraw.Draw(mask)
mask_draw.rectangle([
(position[0] + overlap, position[1] + overlap),
(position[0] + source.width - overlap, position[1] + source.height - overlap)
], fill=0)
# Prepare the image for ControlNet
cnet_image = background.copy()
cnet_image.paste(0, (0, 0), mask)
for image in pipe(
prompt_embeds=prompt_embeds,
negative_prompt_embeds=negative_prompt_embeds,
pooled_prompt_embeds=pooled_prompt_embeds,
negative_pooled_prompt_embeds=negative_pooled_prompt_embeds,
image=cnet_image,
):
yield image, cnet_image
image = image.convert("RGBA")
cnet_image.paste(image, (0, 0), mask)
yield background, cnet_image
"""
@spaces.GPU
def fill_image(image, model_selection):
source = image
target_ratio=(9, 16)
target_height=1280
overlap=48
fade_width=24
# Calculate target dimensions
target_width = (target_height * target_ratio[0]) // target_ratio[1]
# Resize the source image to fit within the target dimensions while maintaining aspect ratio
source_aspect = source.width / source.height
target_aspect = target_width / target_height
if source_aspect > target_aspect:
# Image is wider than target ratio, fit to width
new_width = target_width
new_height = int(new_width / source_aspect)
else:
# Image is taller than target ratio, fit to height
new_height = target_height
new_width = int(new_height * source_aspect)
resized_source = source.resize((new_width, new_height), Image.LANCZOS)
# Calculate margins
margin_x = (target_width - new_width) // 2
margin_y = (target_height - new_height) // 2
# Create a white background
background = Image.new('RGB', (target_width, target_height), (255, 255, 255))
# Paste the resized image onto the white background
position = (margin_x, margin_y)
background.paste(resized_source, position)
# Create the mask with gradient edges
mask = Image.new('L', (target_width, target_height), 255)
mask_array = np.array(mask)
# Create gradient for left and right edges
for i in range(fade_width):
alpha = i / fade_width
mask_array[:, margin_x+overlap+i] = np.minimum(mask_array[:, margin_x+overlap+i], int(255 * (1 - alpha)))
mask_array[:, margin_x+new_width-overlap-i-1] = np.minimum(mask_array[:, margin_x+new_width-overlap-i-1], int(255 * (1 - alpha)))
# Create gradient for top and bottom edges
for i in range(fade_width):
alpha = i / fade_width
mask_array[margin_y+overlap+i, :] = np.minimum(mask_array[margin_y+overlap+i, :], int(255 * (1 - alpha)))
mask_array[margin_y+new_height-overlap-i-1, :] = np.minimum(mask_array[margin_y+new_height-overlap-i-1, :], int(255 * (1 - alpha)))
# Set the center to black
mask_array[margin_y+overlap+fade_width:margin_y+new_height-overlap-fade_width,
margin_x+overlap+fade_width:margin_x+new_width-overlap-fade_width] = 0
mask = Image.fromarray(mask_array.astype('uint8'), 'L')
# Prepare the image for ControlNet
cnet_image = background.copy()
cnet_image.paste(0, (0, 0), mask)
for image in pipe(
prompt_embeds=prompt_embeds,
negative_prompt_embeds=negative_prompt_embeds,
pooled_prompt_embeds=pooled_prompt_embeds,
negative_pooled_prompt_embeds=negative_pooled_prompt_embeds,
image=cnet_image,
):
yield image, cnet_image
image = image.convert("RGBA")
cnet_image.paste(image, (0, 0), mask)
yield background, cnet_image
def clear_result():
return gr.update(value=None)
css = """
.gradio-container {
width: 1024px !important;
}
"""
title = """<h1 align="center">Diffusers Image Fill</h1>
<div align="center">Draw the mask over the subject you want to erase or change.</div>
"""
with gr.Blocks(css=css) as demo:
gr.HTML(title)
run_button = gr.Button("Generate")
with gr.Row():
input_image = gr.Image(
type="pil",
label="Input Image",
sources=["upload"],
)
result = ImageSlider(
interactive=False,
label="Generated Image",
)
model_selection = gr.Dropdown(
choices=list(MODELS.keys()),
value="RealVisXL V5.0 Lightning",
label="Model",
)
run_button.click(
fn=clear_result,
inputs=None,
outputs=result,
).then(
fn=fill_image,
inputs=[input_image, model_selection],
outputs=result,
)
demo.launch(share=False)
|