Commit
·
ae9bbd0
1
Parent(s):
807fb92
Refactor to use Hugging Face Inference API with fal-ai provider - Replace local model loading with InferenceClient API - Remove heavy SDXL/ControlNet/BLIP model dependencies - Use FLUX.1-Kontext-dev model via API - Keep FastAPI and Firebase authentication - Significantly reduce memory usage (no local models)
Browse files- app/config.py +3 -1
- app/main_sdxl.py +69 -171
app/config.py
CHANGED
|
@@ -44,8 +44,10 @@ class Settings(BaseSettings):
|
|
| 44 |
"FASTAI_OUTPUT_CAPTION",
|
| 45 |
"Colorized using GAN-Colorization-Model"
|
| 46 |
)
|
| 47 |
-
INFERENCE_PROVIDER: str = os.getenv("INFERENCE_PROVIDER", "
|
|
|
|
| 48 |
INFERENCE_TIMEOUT: int = int(os.getenv("INFERENCE_TIMEOUT", "180"))
|
|
|
|
| 49 |
|
| 50 |
# Storage settings
|
| 51 |
UPLOAD_DIR: str = os.getenv("UPLOAD_DIR", "uploads")
|
|
|
|
| 44 |
"FASTAI_OUTPUT_CAPTION",
|
| 45 |
"Colorized using GAN-Colorization-Model"
|
| 46 |
)
|
| 47 |
+
INFERENCE_PROVIDER: str = os.getenv("INFERENCE_PROVIDER", "fal-ai")
|
| 48 |
+
INFERENCE_MODEL: str = os.getenv("INFERENCE_MODEL", "black-forest-labs/FLUX.1-Kontext-dev")
|
| 49 |
INFERENCE_TIMEOUT: int = int(os.getenv("INFERENCE_TIMEOUT", "180"))
|
| 50 |
+
HF_TOKEN: str = os.getenv("HF_TOKEN", "")
|
| 51 |
|
| 52 |
# Storage settings
|
| 53 |
UPLOAD_DIR: str = os.getenv("UPLOAD_DIR", "uploads")
|
app/main_sdxl.py
CHANGED
|
@@ -1,17 +1,8 @@
|
|
| 1 |
"""
|
| 2 |
-
FastAPI application for Text-Guided Image Colorization using
|
| 3 |
-
|
| 4 |
"""
|
| 5 |
import os
|
| 6 |
-
# Set environment variables BEFORE any imports
|
| 7 |
-
os.environ["OMP_NUM_THREADS"] = "1"
|
| 8 |
-
os.environ["HF_HOME"] = "/tmp/hf_cache"
|
| 9 |
-
os.environ["TRANSFORMERS_CACHE"] = "/tmp/hf_cache"
|
| 10 |
-
os.environ["HF_HUB_CACHE"] = "/tmp/hf_cache"
|
| 11 |
-
os.environ["HUGGINGFACE_HUB_CACHE"] = "/tmp/hf_cache"
|
| 12 |
-
os.environ["XDG_CACHE_HOME"] = "/tmp/hf_cache"
|
| 13 |
-
os.environ["MPLCONFIGDIR"] = "/tmp/matplotlib_config"
|
| 14 |
-
|
| 15 |
import io
|
| 16 |
import uuid
|
| 17 |
import logging
|
|
@@ -25,23 +16,11 @@ from fastapi.staticfiles import StaticFiles
|
|
| 25 |
import firebase_admin
|
| 26 |
from firebase_admin import credentials, app_check, auth as firebase_auth
|
| 27 |
from PIL import Image
|
| 28 |
-
import torch
|
| 29 |
import uvicorn
|
| 30 |
import gradio as gr
|
| 31 |
|
| 32 |
-
#
|
| 33 |
-
from
|
| 34 |
-
from diffusers import (
|
| 35 |
-
AutoencoderKL,
|
| 36 |
-
StableDiffusionXLControlNetPipeline,
|
| 37 |
-
ControlNetModel,
|
| 38 |
-
UNet2DConditionModel,
|
| 39 |
-
)
|
| 40 |
-
from transformers import (
|
| 41 |
-
BlipProcessor, BlipForConditionalGeneration,
|
| 42 |
-
)
|
| 43 |
-
from safetensors.torch import load_file
|
| 44 |
-
from huggingface_hub import hf_hub_download, snapshot_download
|
| 45 |
|
| 46 |
from app.config import settings
|
| 47 |
|
|
@@ -102,12 +81,8 @@ RESULT_DIR = Path("/tmp/colorize_results")
|
|
| 102 |
app.mount("/results", StaticFiles(directory=str(RESULT_DIR)), name="results")
|
| 103 |
app.mount("/uploads", StaticFiles(directory=str(UPLOAD_DIR)), name="uploads")
|
| 104 |
|
| 105 |
-
# Global
|
| 106 |
-
|
| 107 |
-
caption_model = None
|
| 108 |
-
processor = None
|
| 109 |
-
device = None
|
| 110 |
-
weight_dtype = None
|
| 111 |
model_load_error: Optional[str] = None
|
| 112 |
|
| 113 |
# ========== Utility Functions ==========
|
|
@@ -177,110 +152,29 @@ def remove_unlikely_words(prompt: str) -> str:
|
|
| 177 |
|
| 178 |
@app.on_event("startup")
|
| 179 |
async def startup_event():
|
| 180 |
-
"""
|
| 181 |
-
global
|
| 182 |
|
| 183 |
try:
|
| 184 |
-
logger.info("🔄
|
| 185 |
-
|
| 186 |
-
# Use writable directory for model downloads
|
| 187 |
-
controlnet_dir = "/tmp/sdxl_light_caption_output"
|
| 188 |
-
try:
|
| 189 |
-
os.makedirs(controlnet_dir, exist_ok=True)
|
| 190 |
-
# Test write permissions
|
| 191 |
-
test_file = os.path.join(controlnet_dir, ".test_write")
|
| 192 |
-
with open(test_file, "w") as f:
|
| 193 |
-
f.write("test")
|
| 194 |
-
os.remove(test_file)
|
| 195 |
-
logger.info(f"Using directory: {controlnet_dir}")
|
| 196 |
-
except PermissionError as e:
|
| 197 |
-
logger.error(f"Permission denied for directory {controlnet_dir}: {e}")
|
| 198 |
-
raise
|
| 199 |
-
except Exception as e:
|
| 200 |
-
logger.error(f"Failed to create directory {controlnet_dir}: {e}")
|
| 201 |
-
raise
|
| 202 |
-
|
| 203 |
-
# Download controlnet model snapshot
|
| 204 |
-
controlnet_path = os.path.join(controlnet_dir, "checkpoint-30000", "controlnet")
|
| 205 |
-
if os.path.exists(controlnet_path):
|
| 206 |
-
logger.info(f"ControlNet model already exists at {controlnet_path}")
|
| 207 |
-
else:
|
| 208 |
-
try:
|
| 209 |
-
logger.info("Downloading ControlNet model...")
|
| 210 |
-
snapshot_download(
|
| 211 |
-
repo_id='nickpai/sdxl_light_caption_output',
|
| 212 |
-
local_dir=controlnet_dir
|
| 213 |
-
)
|
| 214 |
-
logger.info("ControlNet model downloaded successfully")
|
| 215 |
-
except Exception as e:
|
| 216 |
-
logger.error(f"Could not download controlnet snapshot: {e}")
|
| 217 |
-
if not os.path.exists(controlnet_path):
|
| 218 |
-
raise
|
| 219 |
-
|
| 220 |
-
# Device and precision setup
|
| 221 |
-
accelerator = Accelerator(mixed_precision="fp16")
|
| 222 |
-
weight_dtype = torch.float16 if accelerator.mixed_precision == "fp16" else torch.float32
|
| 223 |
-
device = accelerator.device
|
| 224 |
-
|
| 225 |
-
logger.info(f"Using device: {device}, dtype: {weight_dtype}")
|
| 226 |
-
|
| 227 |
-
# Pretrained paths
|
| 228 |
-
base_model_path = settings.BASE_MODEL_ID
|
| 229 |
-
safetensors_ckpt = settings.LIGHTNING_WEIGHTS
|
| 230 |
-
# controlnet_path already defined above
|
| 231 |
|
| 232 |
-
#
|
| 233 |
-
|
| 234 |
-
|
| 235 |
-
|
| 236 |
-
vae.enable_slicing()
|
| 237 |
-
vae.enable_tiling()
|
| 238 |
|
| 239 |
-
|
| 240 |
-
|
| 241 |
-
|
| 242 |
-
|
| 243 |
-
unet.set_attention_slice("max")
|
| 244 |
-
|
| 245 |
-
logger.info("Loading ControlNet...")
|
| 246 |
-
controlnet = ControlNetModel.from_pretrained(controlnet_path, torch_dtype=weight_dtype)
|
| 247 |
-
# Enable attention slicing for ControlNet
|
| 248 |
-
controlnet.set_attention_slice("max")
|
| 249 |
-
|
| 250 |
-
logger.info("Creating pipeline...")
|
| 251 |
-
pipe = StableDiffusionXLControlNetPipeline.from_pretrained(
|
| 252 |
-
base_model_path, vae=vae, unet=unet, controlnet=controlnet, torch_dtype=weight_dtype
|
| 253 |
)
|
| 254 |
-
pipe.safety_checker = None
|
| 255 |
-
|
| 256 |
-
# Enable sequential CPU offloading to reduce memory usage
|
| 257 |
-
logger.info("Enabling CPU offloading for memory efficiency...")
|
| 258 |
-
pipe.enable_sequential_cpu_offload()
|
| 259 |
-
# Alternative: use model CPU offload (moves entire model to CPU when not in use)
|
| 260 |
-
# pipe.enable_model_cpu_offload()
|
| 261 |
|
| 262 |
-
logger.info("
|
| 263 |
-
|
| 264 |
-
# Load BLIP captioning model (use base to save memory)
|
| 265 |
-
logger.info("Loading BLIP captioning model (using base model for memory efficiency)...")
|
| 266 |
-
caption_model_name = "blip-image-captioning-base"
|
| 267 |
-
try:
|
| 268 |
-
processor = BlipProcessor.from_pretrained(f"Salesforce/{caption_model_name}")
|
| 269 |
-
caption_model = BlipForConditionalGeneration.from_pretrained(
|
| 270 |
-
f"Salesforce/{caption_model_name}", torch_dtype=weight_dtype
|
| 271 |
-
)
|
| 272 |
-
# Keep BLIP on CPU and move to device only during inference
|
| 273 |
-
caption_model.eval()
|
| 274 |
-
except Exception as e:
|
| 275 |
-
logger.error(f"Failed to load BLIP model: {e}")
|
| 276 |
-
raise
|
| 277 |
-
|
| 278 |
-
logger.info("✅ All models loaded successfully!")
|
| 279 |
model_load_error = None
|
| 280 |
|
| 281 |
except Exception as e:
|
| 282 |
error_msg = str(e)
|
| 283 |
-
logger.error(f"❌ Failed to
|
| 284 |
model_load_error = error_msg
|
| 285 |
# Don't raise - allow health check to work
|
| 286 |
|
|
@@ -288,11 +182,9 @@ async def startup_event():
|
|
| 288 |
@app.on_event("shutdown")
|
| 289 |
async def shutdown_event():
|
| 290 |
"""Cleanup on shutdown"""
|
| 291 |
-
global
|
| 292 |
-
if
|
| 293 |
-
|
| 294 |
-
if caption_model:
|
| 295 |
-
del caption_model
|
| 296 |
logger.info("Application shutdown")
|
| 297 |
|
| 298 |
|
|
@@ -356,9 +248,9 @@ async def health_check():
|
|
| 356 |
"""Health check endpoint"""
|
| 357 |
response = {
|
| 358 |
"status": "healthy",
|
| 359 |
-
"model_loaded":
|
| 360 |
-
"model_type": "
|
| 361 |
-
"
|
| 362 |
}
|
| 363 |
if model_load_error:
|
| 364 |
response["model_error"] = model_load_error
|
|
@@ -373,7 +265,7 @@ def colorize_image_sdxl(
|
|
| 373 |
num_inference_steps: int = 8
|
| 374 |
) -> Tuple[Image.Image, str]:
|
| 375 |
"""
|
| 376 |
-
Colorize a grayscale or low-color image using
|
| 377 |
|
| 378 |
Args:
|
| 379 |
image: PIL Image to colorize
|
|
@@ -385,46 +277,52 @@ def colorize_image_sdxl(
|
|
| 385 |
Returns:
|
| 386 |
Tuple of (colorized PIL Image, caption string)
|
| 387 |
"""
|
| 388 |
-
if
|
| 389 |
-
raise RuntimeError("
|
| 390 |
|
| 391 |
-
torch.manual_seed(seed)
|
| 392 |
original_size = image.size
|
| 393 |
-
|
|
|
|
| 394 |
|
| 395 |
-
#
|
| 396 |
-
|
| 397 |
-
|
| 398 |
-
|
| 399 |
-
|
| 400 |
-
with torch.no_grad():
|
| 401 |
-
caption_ids = caption_model.generate(**inputs, max_length=50, num_beams=3)
|
| 402 |
-
caption = processor.decode(caption_ids[0], skip_special_tokens=True)
|
| 403 |
-
caption = remove_unlikely_words(caption)
|
| 404 |
|
| 405 |
-
# Construct
|
| 406 |
-
|
| 407 |
-
|
|
|
|
|
|
|
| 408 |
else:
|
| 409 |
-
final_prompt =
|
| 410 |
|
| 411 |
-
# Inference
|
| 412 |
-
|
| 413 |
-
|
|
|
|
|
|
|
|
|
|
| 414 |
prompt=final_prompt,
|
| 415 |
-
|
| 416 |
-
num_inference_steps=num_inference_steps,
|
| 417 |
-
generator=torch.manual_seed(seed),
|
| 418 |
-
image=control_image,
|
| 419 |
-
guidance_scale=7.5, # Lower guidance scale uses less memory
|
| 420 |
)
|
| 421 |
-
|
| 422 |
-
|
| 423 |
-
|
| 424 |
-
|
| 425 |
-
|
| 426 |
-
|
| 427 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 428 |
|
| 429 |
|
| 430 |
@app.post("/colorize")
|
|
@@ -440,8 +338,8 @@ async def colorize_api(
|
|
| 440 |
Upload a grayscale image -> returns colorized image.
|
| 441 |
Uses SDXL + ControlNet with automatic captioning.
|
| 442 |
"""
|
| 443 |
-
if
|
| 444 |
-
raise HTTPException(status_code=503, detail="
|
| 445 |
|
| 446 |
if not file.content_type or not file.content_type.startswith("image/"):
|
| 447 |
raise HTTPException(status_code=400, detail="File must be an image")
|
|
@@ -505,8 +403,8 @@ def gradio_colorize(image, positive_prompt=None, negative_prompt=None, seed=123)
|
|
| 505 |
if image is None:
|
| 506 |
return None, ""
|
| 507 |
try:
|
| 508 |
-
if
|
| 509 |
-
return None, "
|
| 510 |
colorized, caption = colorize_image_sdxl(
|
| 511 |
image,
|
| 512 |
positive_prompt=positive_prompt,
|
|
@@ -520,7 +418,7 @@ def gradio_colorize(image, positive_prompt=None, negative_prompt=None, seed=123)
|
|
| 520 |
|
| 521 |
|
| 522 |
title = "🎨 Text-Guided Image Colorization"
|
| 523 |
-
description = "Upload a grayscale image and generate a color version
|
| 524 |
|
| 525 |
iface = gr.Interface(
|
| 526 |
fn=gradio_colorize,
|
|
|
|
| 1 |
"""
|
| 2 |
+
FastAPI application for Text-Guided Image Colorization using Hugging Face Inference API
|
| 3 |
+
Uses fal-ai provider for memory-efficient inference
|
| 4 |
"""
|
| 5 |
import os
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 6 |
import io
|
| 7 |
import uuid
|
| 8 |
import logging
|
|
|
|
| 16 |
import firebase_admin
|
| 17 |
from firebase_admin import credentials, app_check, auth as firebase_auth
|
| 18 |
from PIL import Image
|
|
|
|
| 19 |
import uvicorn
|
| 20 |
import gradio as gr
|
| 21 |
|
| 22 |
+
# Hugging Face Inference API
|
| 23 |
+
from huggingface_hub import InferenceClient
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 24 |
|
| 25 |
from app.config import settings
|
| 26 |
|
|
|
|
| 81 |
app.mount("/results", StaticFiles(directory=str(RESULT_DIR)), name="results")
|
| 82 |
app.mount("/uploads", StaticFiles(directory=str(UPLOAD_DIR)), name="uploads")
|
| 83 |
|
| 84 |
+
# Global Inference API client
|
| 85 |
+
inference_client = None
|
|
|
|
|
|
|
|
|
|
|
|
|
| 86 |
model_load_error: Optional[str] = None
|
| 87 |
|
| 88 |
# ========== Utility Functions ==========
|
|
|
|
| 152 |
|
| 153 |
@app.on_event("startup")
|
| 154 |
async def startup_event():
|
| 155 |
+
"""Initialize Hugging Face Inference API client"""
|
| 156 |
+
global inference_client, model_load_error
|
| 157 |
|
| 158 |
try:
|
| 159 |
+
logger.info("🔄 Initializing Hugging Face Inference API client...")
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 160 |
|
| 161 |
+
# Get HF token from environment or settings
|
| 162 |
+
hf_token = os.getenv("HF_TOKEN") or settings.HF_TOKEN
|
| 163 |
+
if not hf_token:
|
| 164 |
+
raise ValueError("HF_TOKEN environment variable is required for Inference API")
|
|
|
|
|
|
|
| 165 |
|
| 166 |
+
# Initialize InferenceClient with fal-ai provider
|
| 167 |
+
inference_client = InferenceClient(
|
| 168 |
+
provider="fal-ai",
|
| 169 |
+
api_key=hf_token,
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 170 |
)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 171 |
|
| 172 |
+
logger.info("✅ Inference API client initialized successfully!")
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 173 |
model_load_error = None
|
| 174 |
|
| 175 |
except Exception as e:
|
| 176 |
error_msg = str(e)
|
| 177 |
+
logger.error(f"❌ Failed to initialize Inference API client: {error_msg}")
|
| 178 |
model_load_error = error_msg
|
| 179 |
# Don't raise - allow health check to work
|
| 180 |
|
|
|
|
| 182 |
@app.on_event("shutdown")
|
| 183 |
async def shutdown_event():
|
| 184 |
"""Cleanup on shutdown"""
|
| 185 |
+
global inference_client
|
| 186 |
+
if inference_client:
|
| 187 |
+
inference_client = None
|
|
|
|
|
|
|
| 188 |
logger.info("Application shutdown")
|
| 189 |
|
| 190 |
|
|
|
|
| 248 |
"""Health check endpoint"""
|
| 249 |
response = {
|
| 250 |
"status": "healthy",
|
| 251 |
+
"model_loaded": inference_client is not None,
|
| 252 |
+
"model_type": "hf_inference_api",
|
| 253 |
+
"provider": "fal-ai"
|
| 254 |
}
|
| 255 |
if model_load_error:
|
| 256 |
response["model_error"] = model_load_error
|
|
|
|
| 265 |
num_inference_steps: int = 8
|
| 266 |
) -> Tuple[Image.Image, str]:
|
| 267 |
"""
|
| 268 |
+
Colorize a grayscale or low-color image using Hugging Face Inference API.
|
| 269 |
|
| 270 |
Args:
|
| 271 |
image: PIL Image to colorize
|
|
|
|
| 277 |
Returns:
|
| 278 |
Tuple of (colorized PIL Image, caption string)
|
| 279 |
"""
|
| 280 |
+
if inference_client is None:
|
| 281 |
+
raise RuntimeError("Inference API client not initialized")
|
| 282 |
|
|
|
|
| 283 |
original_size = image.size
|
| 284 |
+
# Resize to 512x512 for inference (FLUX models work well at this size)
|
| 285 |
+
control_image = image.convert("RGB").resize((512, 512))
|
| 286 |
|
| 287 |
+
# Convert image to bytes for API
|
| 288 |
+
img_bytes = io.BytesIO()
|
| 289 |
+
control_image.save(img_bytes, format="PNG")
|
| 290 |
+
img_bytes.seek(0)
|
| 291 |
+
input_image = img_bytes.read()
|
|
|
|
|
|
|
|
|
|
|
|
|
| 292 |
|
| 293 |
+
# Construct prompt
|
| 294 |
+
base_prompt = positive_prompt or "colorize this image with vibrant natural colors, high quality"
|
| 295 |
+
if negative_prompt:
|
| 296 |
+
# Note: Some models may not support negative_prompt directly
|
| 297 |
+
final_prompt = f"{base_prompt}. Avoid: {negative_prompt}"
|
| 298 |
else:
|
| 299 |
+
final_prompt = base_prompt
|
| 300 |
|
| 301 |
+
# Use Inference API for image-to-image generation
|
| 302 |
+
model_name = settings.INFERENCE_MODEL
|
| 303 |
+
logger.info(f"Calling Inference API with model {model_name}, prompt: {final_prompt}")
|
| 304 |
+
try:
|
| 305 |
+
result_image = inference_client.image_to_image(
|
| 306 |
+
input_image,
|
| 307 |
prompt=final_prompt,
|
| 308 |
+
model=model_name,
|
|
|
|
|
|
|
|
|
|
|
|
|
| 309 |
)
|
| 310 |
+
|
| 311 |
+
# Resize back to original size
|
| 312 |
+
if isinstance(result_image, Image.Image):
|
| 313 |
+
colorized = result_image.resize(original_size)
|
| 314 |
+
else:
|
| 315 |
+
# If it's bytes, convert to PIL Image
|
| 316 |
+
colorized = Image.open(io.BytesIO(result_image)).resize(original_size)
|
| 317 |
+
|
| 318 |
+
# Generate a simple caption from the prompt
|
| 319 |
+
caption = final_prompt[:100] # Truncate for display
|
| 320 |
+
|
| 321 |
+
return colorized, caption
|
| 322 |
+
|
| 323 |
+
except Exception as e:
|
| 324 |
+
logger.error(f"Inference API error: {e}")
|
| 325 |
+
raise RuntimeError(f"Failed to colorize image: {str(e)}")
|
| 326 |
|
| 327 |
|
| 328 |
@app.post("/colorize")
|
|
|
|
| 338 |
Upload a grayscale image -> returns colorized image.
|
| 339 |
Uses SDXL + ControlNet with automatic captioning.
|
| 340 |
"""
|
| 341 |
+
if inference_client is None:
|
| 342 |
+
raise HTTPException(status_code=503, detail="Inference API client not initialized")
|
| 343 |
|
| 344 |
if not file.content_type or not file.content_type.startswith("image/"):
|
| 345 |
raise HTTPException(status_code=400, detail="File must be an image")
|
|
|
|
| 403 |
if image is None:
|
| 404 |
return None, ""
|
| 405 |
try:
|
| 406 |
+
if inference_client is None:
|
| 407 |
+
return None, "Inference API client not initialized"
|
| 408 |
colorized, caption = colorize_image_sdxl(
|
| 409 |
image,
|
| 410 |
positive_prompt=positive_prompt,
|
|
|
|
| 418 |
|
| 419 |
|
| 420 |
title = "🎨 Text-Guided Image Colorization"
|
| 421 |
+
description = "Upload a grayscale image and generate a color version using Hugging Face Inference API (fal-ai provider)."
|
| 422 |
|
| 423 |
iface = gr.Interface(
|
| 424 |
fn=gradio_colorize,
|