Spaces:
Running
on
Zero
Running
on
Zero
Create app.py
Browse files
app.py
ADDED
@@ -0,0 +1,394 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import os
|
2 |
+
import random
|
3 |
+
import sys
|
4 |
+
from typing import Sequence, Mapping, Any, Union
|
5 |
+
import torch
|
6 |
+
import gradio as gr
|
7 |
+
from huggingface_hub import hf_hub_download
|
8 |
+
import spaces
|
9 |
+
|
10 |
+
# Download required models from Hugging Face
|
11 |
+
hf_hub_download(repo_id="black-forest-labs/FLUX.1-dev", filename="ae.safetensors", local_dir="models/vae")
|
12 |
+
hf_hub_download(repo_id="black-forest-labs/FLUX.1-dev", filename="flux1-dev.safetensors", local_dir="models/diffusion_models")
|
13 |
+
hf_hub_download(repo_id="comfyanonymous/flux_text_encoders", filename="clip_l.safetensors", local_dir="models/text_encoders")
|
14 |
+
hf_hub_download(repo_id="comfyanonymous/flux_text_encoders", filename="t5xxl_fp16.safetensors", local_dir="models/text_encoders")
|
15 |
+
hf_hub_download(repo_id="kim2091/UltraSharp", filename="4x-UltraSharp.pth", local_dir="models/upscale_models")
|
16 |
+
|
17 |
+
def get_value_at_index(obj: Union[Sequence, Mapping], index: int) -> Any:
|
18 |
+
"""Returns the value at the given index of a sequence or mapping."""
|
19 |
+
try:
|
20 |
+
return obj[index]
|
21 |
+
except KeyError:
|
22 |
+
return obj["result"][index]
|
23 |
+
|
24 |
+
def find_path(name: str, path: str = None) -> str:
|
25 |
+
"""Recursively looks at parent folders starting from the given path until it finds the given name."""
|
26 |
+
if path is None:
|
27 |
+
path = os.getcwd()
|
28 |
+
|
29 |
+
if name in os.listdir(path):
|
30 |
+
path_name = os.path.join(path, name)
|
31 |
+
print(f"{name} found: {path_name}")
|
32 |
+
return path_name
|
33 |
+
|
34 |
+
parent_directory = os.path.dirname(path)
|
35 |
+
if parent_directory == path:
|
36 |
+
return None
|
37 |
+
|
38 |
+
return find_path(name, parent_directory)
|
39 |
+
|
40 |
+
def add_comfyui_directory_to_sys_path() -> None:
|
41 |
+
"""Add 'ComfyUI' to the sys.path"""
|
42 |
+
comfyui_path = find_path("ComfyUI")
|
43 |
+
if comfyui_path is not None and os.path.isdir(comfyui_path):
|
44 |
+
sys.path.append(comfyui_path)
|
45 |
+
print(f"'{comfyui_path}' added to sys.path")
|
46 |
+
|
47 |
+
def add_extra_model_paths() -> None:
|
48 |
+
"""Parse the optional extra_model_paths.yaml file and add the parsed paths to the sys.path."""
|
49 |
+
try:
|
50 |
+
from main import load_extra_path_config
|
51 |
+
except ImportError:
|
52 |
+
print("Could not import load_extra_path_config from main.py. Looking in utils.extra_config instead.")
|
53 |
+
from utils.extra_config import load_extra_path_config
|
54 |
+
|
55 |
+
extra_model_paths = find_path("extra_model_paths.yaml")
|
56 |
+
if extra_model_paths is not None:
|
57 |
+
load_extra_path_config(extra_model_paths)
|
58 |
+
else:
|
59 |
+
print("Could not find the extra_model_paths config file.")
|
60 |
+
|
61 |
+
add_comfyui_directory_to_sys_path()
|
62 |
+
add_extra_model_paths()
|
63 |
+
|
64 |
+
def import_custom_nodes() -> None:
|
65 |
+
"""Find all custom nodes in the custom_nodes folder and add those node objects to NODE_CLASS_MAPPINGS"""
|
66 |
+
import asyncio
|
67 |
+
import execution
|
68 |
+
from nodes import init_extra_nodes
|
69 |
+
import server
|
70 |
+
|
71 |
+
loop = asyncio.new_event_loop()
|
72 |
+
asyncio.set_event_loop(loop)
|
73 |
+
|
74 |
+
server_instance = server.PromptServer(loop)
|
75 |
+
execution.PromptQueue(server_instance)
|
76 |
+
init_extra_nodes()
|
77 |
+
|
78 |
+
from nodes import NODE_CLASS_MAPPINGS
|
79 |
+
|
80 |
+
# Pre-load models outside the decorated function for ZeroGPU efficiency
|
81 |
+
import_custom_nodes()
|
82 |
+
|
83 |
+
# Initialize model loaders
|
84 |
+
dualcliploader = NODE_CLASS_MAPPINGS["DualCLIPLoader"]()
|
85 |
+
dualcliploader_54 = dualcliploader.load_clip(
|
86 |
+
clip_name1="clip_l.safetensors",
|
87 |
+
clip_name2="t5xxl_fp16.safetensors",
|
88 |
+
type="flux",
|
89 |
+
device="default",
|
90 |
+
)
|
91 |
+
|
92 |
+
upscalemodelloader = NODE_CLASS_MAPPINGS["UpscaleModelLoader"]()
|
93 |
+
upscalemodelloader_44 = upscalemodelloader.load_model(model_name="4x-UltraSharp.pth")
|
94 |
+
|
95 |
+
vaeloader = NODE_CLASS_MAPPINGS["VAELoader"]()
|
96 |
+
vaeloader_55 = vaeloader.load_vae(vae_name="ae.safetensors")
|
97 |
+
|
98 |
+
unetloader = NODE_CLASS_MAPPINGS["UNETLoader"]()
|
99 |
+
unetloader_58 = unetloader.load_unet(
|
100 |
+
unet_name="flux1-dev.safetensors", weight_dtype="default"
|
101 |
+
)
|
102 |
+
|
103 |
+
downloadandloadflorence2model = NODE_CLASS_MAPPINGS["DownloadAndLoadFlorence2Model"]()
|
104 |
+
downloadandloadflorence2model_52 = downloadandloadflorence2model.loadmodel(
|
105 |
+
model="microsoft/Florence-2-large", precision="fp16", attention="sdpa"
|
106 |
+
)
|
107 |
+
|
108 |
+
# Pre-load models to GPU for efficiency
|
109 |
+
from comfy import model_management
|
110 |
+
model_loaders = [dualcliploader_54, vaeloader_55, unetloader_58, downloadandloadflorence2model_52]
|
111 |
+
valid_models = [
|
112 |
+
getattr(loader[0], 'patcher', loader[0])
|
113 |
+
for loader in model_loaders
|
114 |
+
if not isinstance(loader[0], dict) and not isinstance(getattr(loader[0], 'patcher', None), dict)
|
115 |
+
]
|
116 |
+
model_management.load_models_gpu(valid_models)
|
117 |
+
|
118 |
+
@spaces.GPU(duration=120) # Adjust duration based on your workflow speed
|
119 |
+
def enhance_image(image_input, upscale_factor, steps, cfg_scale, denoise_strength, guidance_scale):
|
120 |
+
"""
|
121 |
+
Main function to enhance and upscale images using Florence-2 captioning and FLUX upscaling
|
122 |
+
"""
|
123 |
+
try:
|
124 |
+
with torch.inference_mode():
|
125 |
+
# Handle different input types (file upload vs URL)
|
126 |
+
if isinstance(image_input, str) and image_input.startswith(('http://', 'https://')):
|
127 |
+
# Load from URL
|
128 |
+
load_image_from_url_mtb = NODE_CLASS_MAPPINGS["Load Image From Url (mtb)"]()
|
129 |
+
load_image_result = load_image_from_url_mtb.load(url=image_input)
|
130 |
+
else:
|
131 |
+
# Load from uploaded file
|
132 |
+
loadimage = NODE_CLASS_MAPPINGS["LoadImage"]()
|
133 |
+
load_image_result = loadimage.load_image(image=image_input)
|
134 |
+
|
135 |
+
# Generate detailed caption using Florence-2
|
136 |
+
florence2run = NODE_CLASS_MAPPINGS["Florence2Run"]()
|
137 |
+
florence2run_51 = florence2run.encode(
|
138 |
+
text_input="",
|
139 |
+
task="more_detailed_caption",
|
140 |
+
fill_mask=True,
|
141 |
+
keep_model_loaded=False,
|
142 |
+
max_new_tokens=1024,
|
143 |
+
num_beams=3,
|
144 |
+
do_sample=True,
|
145 |
+
output_mask_select="",
|
146 |
+
seed=random.randint(1, 2**64),
|
147 |
+
image=get_value_at_index(load_image_result, 0),
|
148 |
+
florence2_model=get_value_at_index(downloadandloadflorence2model_52, 0),
|
149 |
+
)
|
150 |
+
|
151 |
+
# Encode the generated caption
|
152 |
+
cliptextencode = NODE_CLASS_MAPPINGS["CLIPTextEncode"]()
|
153 |
+
cliptextencode_6 = cliptextencode.encode(
|
154 |
+
text=get_value_at_index(florence2run_51, 2),
|
155 |
+
clip=get_value_at_index(dualcliploader_54, 0),
|
156 |
+
)
|
157 |
+
|
158 |
+
# Encode empty negative prompt
|
159 |
+
cliptextencode_42 = cliptextencode.encode(
|
160 |
+
text="", clip=get_value_at_index(dualcliploader_54, 0)
|
161 |
+
)
|
162 |
+
|
163 |
+
# Set up upscale factor
|
164 |
+
primitivefloat = NODE_CLASS_MAPPINGS["PrimitiveFloat"]()
|
165 |
+
primitivefloat_60 = primitivefloat.execute(value=upscale_factor)
|
166 |
+
|
167 |
+
# Apply FLUX guidance
|
168 |
+
fluxguidance = NODE_CLASS_MAPPINGS["FluxGuidance"]()
|
169 |
+
fluxguidance_26 = fluxguidance.append(
|
170 |
+
guidance=guidance_scale,
|
171 |
+
conditioning=get_value_at_index(cliptextencode_6, 0)
|
172 |
+
)
|
173 |
+
|
174 |
+
# Perform ultimate upscaling
|
175 |
+
ultimatesdupscale = NODE_CLASS_MAPPINGS["UltimateSDUpscale"]()
|
176 |
+
ultimatesdupscale_50 = ultimatesdupscale.upscale(
|
177 |
+
upscale_by=get_value_at_index(primitivefloat_60, 0),
|
178 |
+
seed=random.randint(1, 2**64),
|
179 |
+
steps=steps,
|
180 |
+
cfg=cfg_scale,
|
181 |
+
sampler_name="euler",
|
182 |
+
scheduler="normal",
|
183 |
+
denoise=denoise_strength,
|
184 |
+
mode_type="Linear",
|
185 |
+
tile_width=1024,
|
186 |
+
tile_height=1024,
|
187 |
+
mask_blur=8,
|
188 |
+
tile_padding=32,
|
189 |
+
seam_fix_mode="None",
|
190 |
+
seam_fix_denoise=1,
|
191 |
+
seam_fix_width=64,
|
192 |
+
seam_fix_mask_blur=8,
|
193 |
+
seam_fix_padding=16,
|
194 |
+
force_uniform_tiles=True,
|
195 |
+
tiled_decode=False,
|
196 |
+
image=get_value_at_index(load_image_result, 0),
|
197 |
+
model=get_value_at_index(unetloader_58, 0),
|
198 |
+
positive=get_value_at_index(fluxguidance_26, 0),
|
199 |
+
negative=get_value_at_index(cliptextencode_42, 0),
|
200 |
+
vae=get_value_at_index(vaeloader_55, 0),
|
201 |
+
upscale_model=get_value_at_index(upscalemodelloader_44, 0),
|
202 |
+
)
|
203 |
+
|
204 |
+
# Save the result
|
205 |
+
saveimage = NODE_CLASS_MAPPINGS["SaveImage"]()
|
206 |
+
saveimage_43 = saveimage.save_images(
|
207 |
+
filename_prefix="enhanced_image",
|
208 |
+
images=get_value_at_index(ultimatesdupscale_50, 0),
|
209 |
+
)
|
210 |
+
|
211 |
+
# Return the path to the saved image
|
212 |
+
saved_path = f"output/{saveimage_43['ui']['images'][0]['filename']}"
|
213 |
+
|
214 |
+
# Also return the generated caption for user feedback
|
215 |
+
generated_caption = get_value_at_index(florence2run_51, 2)
|
216 |
+
|
217 |
+
return saved_path, generated_caption
|
218 |
+
|
219 |
+
except Exception as e:
|
220 |
+
print(f"Error in enhance_image: {str(e)}")
|
221 |
+
raise gr.Error(f"Enhancement failed: {str(e)}")
|
222 |
+
|
223 |
+
# Create the Gradio interface
|
224 |
+
def create_interface():
|
225 |
+
with gr.Blocks(
|
226 |
+
title="π AI Image Enhancer - Florence-2 + FLUX",
|
227 |
+
theme=gr.themes.Soft(),
|
228 |
+
css="""
|
229 |
+
.gradio-container {
|
230 |
+
max-width: 1200px !important;
|
231 |
+
}
|
232 |
+
.main-header {
|
233 |
+
text-align: center;
|
234 |
+
margin-bottom: 2rem;
|
235 |
+
}
|
236 |
+
.result-gallery {
|
237 |
+
min-height: 400px;
|
238 |
+
}
|
239 |
+
"""
|
240 |
+
) as app:
|
241 |
+
|
242 |
+
gr.HTML("""
|
243 |
+
<div class="main-header">
|
244 |
+
<h1>π¨ AI Image Enhancer</h1>
|
245 |
+
<p>Upload an image or provide a URL to enhance it using Florence-2 captioning and FLUX upscaling</p>
|
246 |
+
</div>
|
247 |
+
""")
|
248 |
+
|
249 |
+
with gr.Row():
|
250 |
+
with gr.Column(scale=1):
|
251 |
+
gr.HTML("<h3>π€ Input Settings</h3>")
|
252 |
+
|
253 |
+
with gr.Tabs():
|
254 |
+
with gr.TabItem("π Upload Image"):
|
255 |
+
image_upload = gr.Image(
|
256 |
+
label="Upload Image",
|
257 |
+
type="filepath",
|
258 |
+
height=300
|
259 |
+
)
|
260 |
+
|
261 |
+
with gr.TabItem("π Image URL"):
|
262 |
+
image_url = gr.Textbox(
|
263 |
+
label="Image URL",
|
264 |
+
placeholder="https://example.com/image.jpg",
|
265 |
+
value="https://upload.wikimedia.org/wikipedia/commons/thumb/a/a7/Example.jpg/800px-Example.jpg"
|
266 |
+
)
|
267 |
+
|
268 |
+
gr.HTML("<h3>βοΈ Enhancement Settings</h3>")
|
269 |
+
|
270 |
+
upscale_factor = gr.Slider(
|
271 |
+
minimum=1.0,
|
272 |
+
maximum=4.0,
|
273 |
+
value=2.0,
|
274 |
+
step=0.5,
|
275 |
+
label="Upscale Factor",
|
276 |
+
info="How much to upscale the image"
|
277 |
+
)
|
278 |
+
|
279 |
+
steps = gr.Slider(
|
280 |
+
minimum=10,
|
281 |
+
maximum=50,
|
282 |
+
value=25,
|
283 |
+
step=5,
|
284 |
+
label="Steps",
|
285 |
+
info="Number of denoising steps"
|
286 |
+
)
|
287 |
+
|
288 |
+
cfg_scale = gr.Slider(
|
289 |
+
minimum=0.5,
|
290 |
+
maximum=10.0,
|
291 |
+
value=1.0,
|
292 |
+
step=0.5,
|
293 |
+
label="CFG Scale",
|
294 |
+
info="Classifier-free guidance scale"
|
295 |
+
)
|
296 |
+
|
297 |
+
denoise_strength = gr.Slider(
|
298 |
+
minimum=0.1,
|
299 |
+
maximum=1.0,
|
300 |
+
value=0.3,
|
301 |
+
step=0.1,
|
302 |
+
label="Denoise Strength",
|
303 |
+
info="How much to denoise the image"
|
304 |
+
)
|
305 |
+
|
306 |
+
guidance_scale = gr.Slider(
|
307 |
+
minimum=1.0,
|
308 |
+
maximum=10.0,
|
309 |
+
value=3.5,
|
310 |
+
step=0.5,
|
311 |
+
label="Guidance Scale",
|
312 |
+
info="FLUX guidance strength"
|
313 |
+
)
|
314 |
+
|
315 |
+
enhance_btn = gr.Button(
|
316 |
+
"π Enhance Image",
|
317 |
+
variant="primary",
|
318 |
+
size="lg"
|
319 |
+
)
|
320 |
+
|
321 |
+
with gr.Column(scale=1):
|
322 |
+
gr.HTML("<h3>π Results</h3>")
|
323 |
+
|
324 |
+
output_image = gr.Image(
|
325 |
+
label="Enhanced Image",
|
326 |
+
type="filepath",
|
327 |
+
height=400,
|
328 |
+
interactive=False
|
329 |
+
)
|
330 |
+
|
331 |
+
generated_caption = gr.Textbox(
|
332 |
+
label="Generated Caption",
|
333 |
+
placeholder="The AI-generated caption will appear here...",
|
334 |
+
lines=3,
|
335 |
+
interactive=False
|
336 |
+
)
|
337 |
+
|
338 |
+
gr.HTML("""
|
339 |
+
<div style="margin-top: 1rem; padding: 1rem; background: #f0f0f0; border-radius: 8px;">
|
340 |
+
<h4>π‘ How it works:</h4>
|
341 |
+
<ol>
|
342 |
+
<li>Florence-2 analyzes your image and generates a detailed caption</li>
|
343 |
+
<li>FLUX uses this caption to guide the upscaling process</li>
|
344 |
+
<li>The result is an enhanced, higher-resolution image</li>
|
345 |
+
</ol>
|
346 |
+
</div>
|
347 |
+
""")
|
348 |
+
|
349 |
+
# Event handlers
|
350 |
+
def process_image(img_upload, img_url, upscale_f, steps_val, cfg_val, denoise_val, guidance_val):
|
351 |
+
# Determine input source
|
352 |
+
image_input = img_upload if img_upload is not None else img_url
|
353 |
+
|
354 |
+
if not image_input:
|
355 |
+
raise gr.Error("Please provide an image (upload or URL)")
|
356 |
+
|
357 |
+
return enhance_image(image_input, upscale_f, steps_val, cfg_val, denoise_val, guidance_val)
|
358 |
+
|
359 |
+
enhance_btn.click(
|
360 |
+
fn=process_image,
|
361 |
+
inputs=[
|
362 |
+
image_upload,
|
363 |
+
image_url,
|
364 |
+
upscale_factor,
|
365 |
+
steps,
|
366 |
+
cfg_scale,
|
367 |
+
denoise_strength,
|
368 |
+
guidance_scale
|
369 |
+
],
|
370 |
+
outputs=[output_image, generated_caption]
|
371 |
+
)
|
372 |
+
|
373 |
+
# Example inputs
|
374 |
+
gr.Examples(
|
375 |
+
examples=[
|
376 |
+
[None, "https://upload.wikimedia.org/wikipedia/commons/thumb/a/a7/Example.jpg/800px-Example.jpg", 2.0, 25, 1.0, 0.3, 3.5],
|
377 |
+
[None, "https://picsum.photos/512/512", 2.0, 20, 1.5, 0.4, 4.0],
|
378 |
+
],
|
379 |
+
inputs=[
|
380 |
+
image_upload,
|
381 |
+
image_url,
|
382 |
+
upscale_factor,
|
383 |
+
steps,
|
384 |
+
cfg_scale,
|
385 |
+
denoise_strength,
|
386 |
+
guidance_scale
|
387 |
+
]
|
388 |
+
)
|
389 |
+
|
390 |
+
return app
|
391 |
+
|
392 |
+
if __name__ == "__main__":
|
393 |
+
app = create_interface()
|
394 |
+
app.launch(share=True, server_name="0.0.0.0", server_port=7860)
|