Spaces:
Running
on
Zero
Running
on
Zero
File size: 8,299 Bytes
edb0494 6405936 edb0494 6405936 edb0494 a7d8817 d49f90c a7d8817 6405936 4ab7724 a7d8817 80b786b 4ab7724 80b786b 4ab7724 a7d8817 6405936 4ab7724 6405936 4ab7724 6405936 4ab7724 6405936 a7d8817 6405936 a7d8817 15a8627 5bc9409 4b78a6c 5bc9409 4b78a6c 5bc9409 4b78a6c 6405936 97567b1 40a5fd5 97567b1 9cdaf5d 6405936 97567b1 976671e 0ef5dee 976671e 40a5fd5 4ab7724 40a5fd5 0ef5dee 40a5fd5 0ef5dee 4ab7724 5bc9409 40a5fd5 5bc9409 c837d9c 4ab7724 c837d9c 4ab7724 c837d9c 4ab7724 976671e 97567b1 5bc9409 6405936 976671e 4ab7724 6405936 4ab7724 6405936 4ab7724 |
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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 |
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)
@spaces.GPU
def infer(image, model_selection, width, height, overlap_width, num_inference_steps, prompt_input=None):
source = image
target_size = (width, height)
target_ratio = (width, height) # Calculate aspect ratio from width and height
overlap = overlap_width
# Upscale if source is smaller than target in both dimensions
if source.width < target_size[0] and source.height < target_size[1]:
scale_factor = min(target_size[0] / source.width, target_size[1] / source.height)
new_width = int(source.width * scale_factor)
new_height = int(source.height * scale_factor)
source = source.resize((new_width, new_height), Image.LANCZOS)
if source.width > target_size[0] or source.height > target_size[1]:
scale_factor = min(target_size[0] / source.width, target_size[1] / source.height)
new_width = int(source.width * scale_factor)
new_height = int(source.height * scale_factor)
source = source.resize((new_width, new_height), Image.LANCZOS)
margin_x = (target_size[0] - source.width) // 2
margin_y = (target_size[1] - source.height) // 2
background = Image.new('RGB', target_size, (255, 255, 255))
background.paste(source, (margin_x, margin_y))
mask = Image.new('L', target_size, 255)
mask_draw = ImageDraw.Draw(mask)
mask_draw.rectangle([
(margin_x + overlap, margin_y + overlap),
(margin_x + source.width - overlap, margin_y + source.height - overlap)
], fill=0)
cnet_image = background.copy()
cnet_image.paste(0, (0, 0), mask)
final_prompt = "high quality"
if prompt_input.strip() != "":
final_prompt += ", " + prompt_input
(
prompt_embeds,
negative_prompt_embeds,
pooled_prompt_embeds,
negative_pooled_prompt_embeds,
) = pipe.encode_prompt(final_prompt, "cuda", True)
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,
num_inference_steps=num_inference_steps
):
yield cnet_image, image
image = image.convert("RGBA")
cnet_image.paste(image, (0, 0), mask)
yield background, cnet_image
def preload_presets(target_ratio):
if target_ratio == "9:16":
changed_width = 720
changed_height = 1280
return changed_width, changed_height, gr.update(open=False)
elif target_ratio == "16:9":
changed_width = 1280
changed_height = 720
return changed_width, changed_height, gr.update(open=False)
elif target_ratio == "Custom":
return 720, 1280, gr.update(open=True)
def clear_result():
return gr.update(value=None)
css = """
.gradio-container {
width: 1200px !important;
}
"""
title = """<h1 align="center">Diffusers Image Outpaint</h1>
<div align="center">Drop an image you would like to extend, pick your expected ratio and hit Generate.</div>
"""
with gr.Blocks(css=css) as demo:
with gr.Column():
gr.HTML(title)
with gr.Row():
with gr.Column():
input_image = gr.Image(
type="pil",
label="Input Image",
sources=["upload"],
height = 300
)
prompt_input = gr.Textbox(label="Prompt (Optional)")
with gr.Row():
target_ratio = gr.Radio(
label = "Expected Ratio",
choices = ["9:16", "16:9", "Custom"],
value = "9:16",
scale = 2
)
run_button = gr.Button("Generate", scale=1)
with gr.Accordion(label="Advanced settings", open=False) as settings_panel:
with gr.Column():
with gr.Row():
width_slider = gr.Slider(
label="Width",
minimum=720,
maximum=1440,
step=8,
value=720, # Set a default value
)
height_slider = gr.Slider(
label="Height",
minimum=720,
maximum=1440,
step=8,
value=1024, # Set a default value
)
with gr.Row():
model_selection = gr.Dropdown(
choices=list(MODELS.keys()),
value="RealVisXL V5.0 Lightning",
label="Model",
)
num_inference_steps = gr.Slider(label="Steps", minimum=4, maximum=12, step=1, value=8 )
overlap_width = gr.Slider(
label="Mask overlap width",
minimum=1,
maximum=50,
value=42,
step=1
)
gr.Examples(
examples=[
["./examples/example_1.webp", "RealVisXL V5.0 Lightning", 1280, 720],
["./examples/example_2.jpg", "RealVisXL V5.0 Lightning", 720, 1280],
["./examples/example_3.jpg", "RealVisXL V5.0 Lightning", 1024, 1024],
],
inputs=[input_image, model_selection, width_slider, height_slider],
)
with gr.Column():
result = ImageSlider(
interactive=False,
label="Generated Image",
)
target_ratio.change(
fn = preload_presets,
inputs = [target_ratio],
outputs = [width_slider, height_slider, settings_panel],
queue = False
)
run_button.click(
fn=clear_result,
inputs=None,
outputs=result,
).then(
fn=infer,
inputs=[input_image, model_selection, width_slider, height_slider, overlap_width, num_inference_steps, prompt_input],
outputs=result,
)
prompt_input.submit(
fn=clear_result,
inputs=None,
outputs=result,
).then(
fn=infer,
inputs=[input_image, model_selection, width_slider, height_slider, overlap_width, num_inference_steps, prompt_input],
outputs=result,
)
demo.launch(share=False) |