Spaces:
Runtime error
Runtime error
Create app.py
Browse files
app.py
ADDED
@@ -0,0 +1,352 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import gradio as gr
|
2 |
+
import torch
|
3 |
+
from diffusers import LTXVideoTransformer3DModel, LTXVideoPipeline
|
4 |
+
from transformers import T5EncoderModel, T5Tokenizer
|
5 |
+
import spaces
|
6 |
+
import numpy as np
|
7 |
+
import tempfile
|
8 |
+
import os
|
9 |
+
import time
|
10 |
+
import logging
|
11 |
+
from PIL import Image
|
12 |
+
import cv2
|
13 |
+
from fastapi import FastAPI, File, UploadFile, Form, HTTPException
|
14 |
+
from fastapi.responses import FileResponse
|
15 |
+
import uvicorn
|
16 |
+
import threading
|
17 |
+
import json
|
18 |
+
|
19 |
+
# Configure logging
|
20 |
+
logging.basicConfig(level=logging.INFO)
|
21 |
+
logger = logging.getLogger(__name__)
|
22 |
+
|
23 |
+
# Global variables for model
|
24 |
+
pipe = None
|
25 |
+
device = "cuda" if torch.cuda.is_available() else "cpu"
|
26 |
+
|
27 |
+
def load_model():
|
28 |
+
"""Load the LTX-Video model with optimizations"""
|
29 |
+
global pipe
|
30 |
+
try:
|
31 |
+
logger.info("Loading LTX-Video model...")
|
32 |
+
|
33 |
+
# Load the pipeline
|
34 |
+
pipe = LTXVideoPipeline.from_pretrained(
|
35 |
+
"Lightricks/LTX-Video-0.9.7-dev",
|
36 |
+
torch_dtype=torch.bfloat16,
|
37 |
+
use_safetensors=True
|
38 |
+
)
|
39 |
+
|
40 |
+
# Move to device
|
41 |
+
pipe = pipe.to(device)
|
42 |
+
|
43 |
+
# Enable optimizations
|
44 |
+
pipe.vae.enable_tiling()
|
45 |
+
pipe.vae.enable_slicing()
|
46 |
+
|
47 |
+
# Enable memory efficient attention if available
|
48 |
+
if hasattr(pipe.unet, 'enable_xformers_memory_efficient_attention'):
|
49 |
+
pipe.unet.enable_xformers_memory_efficient_attention()
|
50 |
+
|
51 |
+
logger.info("Model loaded successfully!")
|
52 |
+
return True
|
53 |
+
except Exception as e:
|
54 |
+
logger.error(f"Error loading model: {e}")
|
55 |
+
return False
|
56 |
+
|
57 |
+
def validate_inputs(prompt, duration, image=None):
|
58 |
+
"""Validate input parameters"""
|
59 |
+
errors = []
|
60 |
+
|
61 |
+
if not prompt or len(prompt.strip()) == 0:
|
62 |
+
errors.append("Prompt is required")
|
63 |
+
|
64 |
+
if len(prompt) > 500:
|
65 |
+
errors.append("Prompt must be less than 500 characters")
|
66 |
+
|
67 |
+
if duration < 3 or duration > 5:
|
68 |
+
errors.append("Duration must be between 3 and 5 seconds")
|
69 |
+
|
70 |
+
if image is not None:
|
71 |
+
try:
|
72 |
+
if isinstance(image, str):
|
73 |
+
img = Image.open(image)
|
74 |
+
else:
|
75 |
+
img = image
|
76 |
+
|
77 |
+
# Check image dimensions
|
78 |
+
width, height = img.size
|
79 |
+
if width > 1024 or height > 1024:
|
80 |
+
errors.append("Image dimensions must be less than 1024x1024")
|
81 |
+
|
82 |
+
except Exception as e:
|
83 |
+
errors.append(f"Invalid image: {str(e)}")
|
84 |
+
|
85 |
+
return errors
|
86 |
+
|
87 |
+
def frames_to_video(frames, output_path, fps=24):
|
88 |
+
"""Convert frames to video using OpenCV"""
|
89 |
+
try:
|
90 |
+
height, width = frames[0].shape[:2]
|
91 |
+
fourcc = cv2.VideoWriter_fourcc(*'mp4v')
|
92 |
+
out = cv2.VideoWriter(output_path, fourcc, fps, (width, height))
|
93 |
+
|
94 |
+
for frame in frames:
|
95 |
+
# Convert RGB to BGR for OpenCV
|
96 |
+
frame_bgr = cv2.cvtColor(frame, cv2.COLOR_RGB2BGR)
|
97 |
+
out.write(frame_bgr)
|
98 |
+
|
99 |
+
out.release()
|
100 |
+
return True
|
101 |
+
except Exception as e:
|
102 |
+
logger.error(f"Error creating video: {e}")
|
103 |
+
return False
|
104 |
+
|
105 |
+
@spaces.GPU(duration=60)
|
106 |
+
def generate_video_core(prompt, negative_prompt="", duration=4, image=None):
|
107 |
+
"""Core video generation function with ZeroGPU decorator"""
|
108 |
+
global pipe
|
109 |
+
|
110 |
+
start_time = time.time()
|
111 |
+
|
112 |
+
try:
|
113 |
+
# Calculate number of frames (24 FPS)
|
114 |
+
num_frames = int(duration * 24)
|
115 |
+
|
116 |
+
# Prepare generation parameters
|
117 |
+
generation_kwargs = {
|
118 |
+
"prompt": prompt,
|
119 |
+
"negative_prompt": negative_prompt,
|
120 |
+
"num_frames": num_frames,
|
121 |
+
"height": 512,
|
122 |
+
"width": 768,
|
123 |
+
"num_inference_steps": 30,
|
124 |
+
"guidance_scale": 7.5,
|
125 |
+
"generator": torch.Generator(device=device).manual_seed(42)
|
126 |
+
}
|
127 |
+
|
128 |
+
# Add image if provided
|
129 |
+
if image is not None:
|
130 |
+
if isinstance(image, str):
|
131 |
+
image = Image.open(image)
|
132 |
+
# Resize image to match output dimensions
|
133 |
+
image = image.resize((768, 512), Image.Resampling.LANCZOS)
|
134 |
+
generation_kwargs["image"] = image
|
135 |
+
|
136 |
+
logger.info(f"Starting generation with {num_frames} frames...")
|
137 |
+
|
138 |
+
# Generate video
|
139 |
+
with torch.inference_mode():
|
140 |
+
result = pipe(**generation_kwargs)
|
141 |
+
|
142 |
+
# Get the generated frames
|
143 |
+
frames = result.frames[0] # First (and only) video in batch
|
144 |
+
|
145 |
+
# Convert to numpy arrays if needed
|
146 |
+
if torch.is_tensor(frames):
|
147 |
+
frames = frames.cpu().numpy()
|
148 |
+
|
149 |
+
# Ensure frames are in the right format (0-255 uint8)
|
150 |
+
if frames.dtype != np.uint8:
|
151 |
+
frames = (frames * 255).astype(np.uint8)
|
152 |
+
|
153 |
+
# Create temporary video file
|
154 |
+
temp_dir = tempfile.mkdtemp()
|
155 |
+
video_path = os.path.join(temp_dir, "generated_video.mp4")
|
156 |
+
|
157 |
+
# Convert frames to video
|
158 |
+
success = frames_to_video(frames, video_path, fps=24)
|
159 |
+
|
160 |
+
if not success:
|
161 |
+
raise Exception("Failed to create video file")
|
162 |
+
|
163 |
+
generation_time = time.time() - start_time
|
164 |
+
logger.info(f"Video generated successfully in {generation_time:.2f} seconds")
|
165 |
+
|
166 |
+
return video_path, f"Generated in {generation_time:.2f}s"
|
167 |
+
|
168 |
+
except Exception as e:
|
169 |
+
logger.error(f"Error generating video: {e}")
|
170 |
+
raise Exception(f"Generation failed: {str(e)}")
|
171 |
+
|
172 |
+
def generate_video_gradio(prompt, negative_prompt, duration, image):
|
173 |
+
"""Gradio interface wrapper"""
|
174 |
+
try:
|
175 |
+
# Validate inputs
|
176 |
+
errors = validate_inputs(prompt, duration, image)
|
177 |
+
if errors:
|
178 |
+
return None, f"Validation errors: {'; '.join(errors)}"
|
179 |
+
|
180 |
+
# Check if model is loaded
|
181 |
+
if pipe is None:
|
182 |
+
return None, "Model not loaded. Please wait for initialization."
|
183 |
+
|
184 |
+
# Generate video
|
185 |
+
video_path, status = generate_video_core(prompt, negative_prompt, duration, image)
|
186 |
+
return video_path, status
|
187 |
+
|
188 |
+
except Exception as e:
|
189 |
+
logger.error(f"Gradio generation error: {e}")
|
190 |
+
return None, f"Error: {str(e)}"
|
191 |
+
|
192 |
+
# Create Gradio interface
|
193 |
+
def create_gradio_interface():
|
194 |
+
with gr.Blocks(title="LTX-Video Generator", theme=gr.themes.Soft()) as demo:
|
195 |
+
gr.Markdown("# 🎬 LTX-Video Generator")
|
196 |
+
gr.Markdown("Generate 3-5 second videos using the LTX-Video model from Lightricks")
|
197 |
+
|
198 |
+
with gr.Row():
|
199 |
+
with gr.Column(scale=1):
|
200 |
+
# Input controls
|
201 |
+
image_input = gr.File(
|
202 |
+
label="Input Image (Optional)",
|
203 |
+
file_types=[".png", ".jpg", ".jpeg"],
|
204 |
+
type="filepath"
|
205 |
+
)
|
206 |
+
|
207 |
+
prompt_input = gr.Textbox(
|
208 |
+
label="Prompt",
|
209 |
+
placeholder="Describe the video you want to generate...",
|
210 |
+
lines=3,
|
211 |
+
max_lines=5
|
212 |
+
)
|
213 |
+
|
214 |
+
negative_prompt_input = gr.Textbox(
|
215 |
+
label="Negative Prompt (Optional)",
|
216 |
+
placeholder="What you don't want in the video...",
|
217 |
+
lines=2,
|
218 |
+
max_lines=3
|
219 |
+
)
|
220 |
+
|
221 |
+
duration_slider = gr.Slider(
|
222 |
+
minimum=3,
|
223 |
+
maximum=5,
|
224 |
+
value=4,
|
225 |
+
step=0.5,
|
226 |
+
label="Duration (seconds)"
|
227 |
+
)
|
228 |
+
|
229 |
+
generate_btn = gr.Button("🎬 Generate Video", variant="primary")
|
230 |
+
|
231 |
+
gr.Markdown("**Estimated time:** 4-6 seconds")
|
232 |
+
|
233 |
+
with gr.Column(scale=1):
|
234 |
+
# Output controls
|
235 |
+
video_output = gr.Video(label="Generated Video")
|
236 |
+
status_output = gr.Textbox(label="Status", interactive=False)
|
237 |
+
|
238 |
+
# Event handlers
|
239 |
+
generate_btn.click(
|
240 |
+
fn=generate_video_gradio,
|
241 |
+
inputs=[prompt_input, negative_prompt_input, duration_slider, image_input],
|
242 |
+
outputs=[video_output, status_output]
|
243 |
+
)
|
244 |
+
|
245 |
+
# Examples
|
246 |
+
gr.Examples(
|
247 |
+
examples=[
|
248 |
+
["A cat playing with a ball of yarn", "", 4, None],
|
249 |
+
["Ocean waves crashing on a beach at sunset", "", 3, None],
|
250 |
+
["A person walking through a forest", "blurry, low quality", 5, None],
|
251 |
+
],
|
252 |
+
inputs=[prompt_input, negative_prompt_input, duration_slider, image_input]
|
253 |
+
)
|
254 |
+
|
255 |
+
return demo
|
256 |
+
|
257 |
+
# FastAPI setup
|
258 |
+
app = FastAPI(title="LTX-Video API", description="Generate videos using LTX-Video model")
|
259 |
+
|
260 |
+
@app.post("/generate_video")
|
261 |
+
async def api_generate_video(
|
262 |
+
prompt: str = Form(..., description="Text prompt for video generation"),
|
263 |
+
negative_prompt: str = Form("", description="Negative prompt (optional)"),
|
264 |
+
duration: float = Form(4.0, description="Duration in seconds (3-5)"),
|
265 |
+
image: UploadFile = File(None, description="Input image (optional)")
|
266 |
+
):
|
267 |
+
"""Generate video via API"""
|
268 |
+
try:
|
269 |
+
# Validate inputs
|
270 |
+
image_path = None
|
271 |
+
if image:
|
272 |
+
# Save uploaded image temporarily
|
273 |
+
temp_dir = tempfile.mkdtemp()
|
274 |
+
image_path = os.path.join(temp_dir, image.filename)
|
275 |
+
with open(image_path, "wb") as f:
|
276 |
+
content = await image.read()
|
277 |
+
f.write(content)
|
278 |
+
|
279 |
+
errors = validate_inputs(prompt, duration, image_path)
|
280 |
+
if errors:
|
281 |
+
raise HTTPException(status_code=400, detail={"errors": errors})
|
282 |
+
|
283 |
+
if pipe is None:
|
284 |
+
raise HTTPException(status_code=503, detail="Model not loaded")
|
285 |
+
|
286 |
+
# Generate video
|
287 |
+
video_path, status = generate_video_core(prompt, negative_prompt, duration, image_path)
|
288 |
+
|
289 |
+
# Return video file
|
290 |
+
return FileResponse(
|
291 |
+
video_path,
|
292 |
+
media_type="video/mp4",
|
293 |
+
filename=f"generated_video_{int(time.time())}.mp4"
|
294 |
+
)
|
295 |
+
|
296 |
+
except HTTPException:
|
297 |
+
raise
|
298 |
+
except Exception as e:
|
299 |
+
logger.error(f"API generation error: {e}")
|
300 |
+
raise HTTPException(status_code=500, detail=str(e))
|
301 |
+
|
302 |
+
@app.get("/")
|
303 |
+
async def root():
|
304 |
+
"""API documentation"""
|
305 |
+
return {
|
306 |
+
"message": "LTX-Video API",
|
307 |
+
"endpoints": {
|
308 |
+
"/generate_video": "POST - Generate video",
|
309 |
+
"/docs": "GET - API documentation"
|
310 |
+
},
|
311 |
+
"curl_example": """
|
312 |
+
curl -X POST "http://localhost:7860/generate_video" \\
|
313 |
+
-F "prompt=A cat playing with a ball" \\
|
314 |
+
-F "duration=4" \\
|
315 |
+
-F "negative_prompt=blurry" \\
|
316 |
+
-F "image=@your_image.jpg" \\
|
317 |
+
--output generated_video.mp4
|
318 |
+
"""
|
319 |
+
}
|
320 |
+
|
321 |
+
def run_api():
|
322 |
+
"""Run FastAPI server"""
|
323 |
+
uvicorn.run(app, host="0.0.0.0", port=7861, log_level="info")
|
324 |
+
|
325 |
+
def main():
|
326 |
+
"""Main function"""
|
327 |
+
# Load model
|
328 |
+
logger.info("Initializing LTX-Video Generator...")
|
329 |
+
model_loaded = load_model()
|
330 |
+
|
331 |
+
if not model_loaded:
|
332 |
+
logger.error("Failed to load model. Exiting.")
|
333 |
+
return
|
334 |
+
|
335 |
+
# Create Gradio interface
|
336 |
+
demo = create_gradio_interface()
|
337 |
+
|
338 |
+
# Start API server in a separate thread
|
339 |
+
api_thread = threading.Thread(target=run_api, daemon=True)
|
340 |
+
api_thread.start()
|
341 |
+
logger.info("API server started on http://localhost:7861")
|
342 |
+
|
343 |
+
# Launch Gradio interface
|
344 |
+
demo.launch(
|
345 |
+
server_name="0.0.0.0",
|
346 |
+
server_port=7860,
|
347 |
+
share=False,
|
348 |
+
show_api=False
|
349 |
+
)
|
350 |
+
|
351 |
+
if __name__ == "__main__":
|
352 |
+
main()
|