sora-2 / app.py
akhaliq's picture
akhaliq HF Staff
Update Gradio app with multiple files
0ac76ac verified
raw
history blame
20.1 kB
import gradio as gr
import os
import tempfile
import shutil
from typing import Optional, Tuple, Union
from huggingface_hub import InferenceClient, whoami
from pathlib import Path
# Initialize Hugging Face Inference Client with fal-ai provider
client = InferenceClient(
provider="fal-ai",
api_key=os.environ.get("HF_TOKEN"),
bill_to="huggingface",
)
def verify_pro_status(token: Optional[Union[gr.OAuthToken, str]]) -> bool:
"""Verifies if the user is a Hugging Face PRO user or part of an enterprise org."""
if not token:
return False
if isinstance(token, gr.OAuthToken):
token_str = token.token
elif isinstance(token, str):
token_str = token
else:
return False # Should not happen with correct type hints, but for safety
try:
user_info = whoami(token=token_str)
return (
user_info.get("isPro", False) or
any(org.get("isEnterprise", False) for org in user_info.get("orgs", []))
)
except Exception as e:
print(f"Could not verify user's PRO/Enterprise status: {e}")
return False
def cleanup_temp_files():
"""Clean up old temporary video files to prevent storage overflow."""
try:
temp_dir = tempfile.gettempdir()
# Clean up old .mp4 files in temp directory
for file_path in Path(temp_dir).glob("*.mp4"):
try:
# Remove files older than 5 minutes
if file_path.stat().st_mtime < (os.time.time() - 300):
file_path.unlink(missing_ok=True)
except Exception:
pass # Ignore errors for individual files
except Exception as e:
print(f"Cleanup error: {e}")
def generate_video(
prompt: str,
duration: int = 8, # These are not used by the fal.ai sora-2 model directly, but kept for interface consistency
size: str = "1280x720", # These are not used by the fal.ai sora-2 model directly, but kept for interface consistency
api_key: Optional[str] = None
) -> Tuple[Optional[str], str]:
"""
Generate video using Sora-2 Text-to-Video through Hugging Face Inference API with fal-ai provider.
Returns tuple of (video_path, status_message).
"""
# Clean up old files before generating new ones
cleanup_temp_files()
try:
# Use provided API key or environment variable
if api_key:
temp_client = InferenceClient(
provider="fal-ai",
api_key=api_key,
bill_to="huggingface",
)
else:
temp_client = client
if not os.environ.get("HF_TOKEN") and not api_key:
return None, "❌ Please set HF_TOKEN environment variable or provide an API key."
# Call Sora-2 through Hugging Face Inference API
video_bytes = temp_client.text_to_video(
prompt,
model="akhaliq/sora-2", # Specific model for text-to-video
)
# Save to temporary file with proper cleanup
temp_file = tempfile.NamedTemporaryFile(suffix=".mp4", delete=False)
try:
temp_file.write(video_bytes)
temp_file.flush()
video_path = temp_file.name
finally:
temp_file.close()
status_message = f"βœ… Video generated successfully!"
return video_path, status_message
except Exception as e:
error_msg = f"❌ Error generating video: {str(e)}"
return None, error_msg
def generate_image_to_video(
image_path: str,
prompt: str,
api_key: Optional[str] = None
) -> Tuple[Optional[str], str]:
"""
Generate video using Sora-2 Image-to-Video through Hugging Face Inference API with fal-ai provider.
Returns tuple of (video_path, status_message).
"""
cleanup_temp_files() # Clean up old files
if not image_path:
return None, "❌ Please upload an image."
if not prompt or prompt.strip() == "":
return None, "❌ Please enter a prompt for the video generation."
try:
if api_key:
temp_client = InferenceClient(
provider="fal-ai",
api_key=api_key,
bill_to="huggingface",
)
else:
temp_client = client
if not os.environ.get("HF_TOKEN") and not api_key:
return None, "❌ Please set HF_TOKEN environment variable or provide an API key."
with open(image_path, "rb") as image_file:
input_image_bytes = image_file.read()
video_bytes = temp_client.image_to_video(
input_image_bytes,
prompt=prompt,
model="akhaliq/sora-2-image-to-video", # Specific model for image-to-video
)
temp_file = tempfile.NamedTemporaryFile(suffix=".mp4", delete=False)
try:
temp_file.write(video_bytes)
temp_file.flush()
video_path = temp_file.name
finally:
temp_file.close()
status_message = f"βœ… Video generated successfully from image and prompt!"
return video_path, status_message
except Exception as e:
error_msg = f"❌ Error generating video from image: {str(e)}"
return None, error_msg
def generate_with_pro_auth(
prompt: str,
oauth_token: Optional[gr.OAuthToken] = None # Gradio will auto-inject this based on type hint
) -> Tuple[Optional[str], str]:
"""
Wrapper function that checks if user is PRO before generating text-to-video.
"""
# Check if user is PRO
if not verify_pro_status(oauth_token):
raise gr.Error("Access Denied. This app is exclusively for Hugging Face PRO users. Please subscribe to PRO to use this app.")
if not prompt or prompt.strip() == "":
return None, "❌ Please enter a prompt"
# Use the environment token for API calls (with bill_to="huggingface")
# Don't use the user's OAuth token for the API call
video_path, status = generate_video(
prompt,
duration=8,
size="1280x720",
api_key=None # This will use the environment HF_TOKEN
)
return video_path, status
def generate_image_to_video_with_pro_auth(
image_path: str,
prompt: str,
oauth_token: Optional[gr.OAuthToken] = None # Gradio will auto-inject this based on type hint
) -> Tuple[Optional[str], str]:
"""
Wrapper function that checks if user is PRO before generating image-to-video.
"""
if not verify_pro_status(oauth_token):
raise gr.Error("Access Denied. This app is exclusively for Hugging Face PRO users. Please subscribe to PRO to use this app.")
if not image_path:
return None, "❌ Please upload an image."
if not prompt or prompt.strip() == "":
return None, "❌ Please enter a prompt"
video_path, status = generate_image_to_video(
image_path,
prompt,
api_key=None # This will use the environment HF_TOKEN
)
return video_path, status
def simple_generate(prompt: str) -> Optional[str]:
"""Simplified wrapper for text-to-video examples that only returns video."""
if not prompt or prompt.strip() == "":
return None
video_path, _ = generate_video(prompt, duration=8, size="1280x720", api_key=None)
return video_path
def simple_generate_image_to_video(image_path: str, prompt: str) -> Optional[str]:
"""Simplified wrapper for image-to-video examples that only returns video."""
if not image_path or not prompt or prompt.strip() == "":
return None
video_path, _ = generate_image_to_video(image_path, prompt, api_key=None)
return video_path
def create_ui():
"""Create the Gradio interface with PRO verification."""
css = '''
.logo-dark{display: none}
.dark .logo-dark{display: block !important}
.dark .logo-light{display: none}
#sub_title{margin-top: -20px !important}
.pro-badge{
background: linear-gradient(135deg, #667eea 0%, #764ba2 100%);
color: white;
padding: 4px 12px;
border-radius: 20px;
font-size: 0.9em;
font-weight: bold;
display: inline-block;
margin-left: 8px;
}
'''
with gr.Blocks(title="Sora-2 Text & Image-to-Video Generator", theme=gr.themes.Soft(), css=css) as demo:
gr.HTML("""
<div style="text-align: center; max-width: 800px; margin: 0 auto;">
<h1 style="font-size: 2.5em; margin-bottom: 0.5em;">
🎬 Sora-2 Text & Image-to-Video Generator
<span class="pro-badge">PRO</span>
</h1>
<p style="font-size: 1.1em; color: #666; margin-bottom: 20px;">Generate stunning videos using OpenAI's Sora-2 model</p>
<p id="sub_title" style="font-size: 1em; margin-top: 20px; margin-bottom: 15px;">
<strong>Exclusive access for Hugging Face PRO users.</strong>
<a href="http://huggingface.co/subscribe/pro?source=sora2_video" target="_blank" style="color: #667eea;">Subscribe to PRO β†’</a>
</p>
<p style="font-size: 0.9em; color: #999; margin-top: 15px;">
Built with <a href="https://huggingface.co/spaces/akhaliq/anycoder" target="_blank" style="color: #667eea;">anycoder</a>
</p>
</div>
""")
# Login button for OAuth
gr.LoginButton()
# PRO message for non-PRO users
pro_message = gr.Markdown(visible=False)
# Main interface (hidden by default)
main_interface = gr.Column(visible=False)
with main_interface:
gr.HTML("""
<div style="text-align: center; margin: 20px 0;">
<p style="color: #28a745; font-weight: bold;">✨ Welcome PRO User! You have full access to Sora-2.</p>
</div>
""")
with gr.Tabs() as tab_selector:
with gr.TabItem("Text-to-Video", id=0):
with gr.Row():
with gr.Column(scale=1):
prompt_input_text = gr.Textbox(
label="Enter your text prompt",
placeholder="Describe the video you want to create...",
lines=4
)
with gr.Accordion("Advanced Settings", open=False):
gr.Markdown("*Coming soon: Duration and resolution controls*")
generate_btn_text = gr.Button("πŸŽ₯ Generate Video from Text", variant="primary", size="lg")
with gr.Column(scale=1):
video_output_text = gr.Video(
label="Generated Video",
height=400,
interactive=False,
show_download_button=True
)
status_output_text = gr.Textbox(
label="Status",
interactive=False,
visible=True
)
# Examples section with queue disabled
gr.Examples(
examples=[
"A serene beach at sunset with waves gently rolling onto the shore",
"A butterfly emerging from its chrysalis in slow motion",
"Northern lights dancing across a starry night sky",
"A bustling city street transitioning from day to night in timelapse",
"A close-up of coffee being poured into a cup with steam rising",
"Cherry blossoms falling in slow motion in a Japanese garden"
],
inputs=prompt_input_text,
outputs=video_output_text,
fn=simple_generate, # Examples use simplified function
cache_examples=False,
api_name=False,
show_api=False,
)
# Event handler for generation with queue disabled
generate_btn_text.click(
fn=generate_with_pro_auth,
inputs=[prompt_input_text], # OAuth token is auto-injected by type hint
outputs=[video_output_text, status_output_text],
queue=False,
api_name=False,
show_api=False,
)
with gr.TabItem("Image-to-Video", id=1):
with gr.Row():
with gr.Column(scale=1):
image_input = gr.Image(
label="Upload your input image",
type="filepath",
height=300,
value="https://huggingface.co/spaces/akhaliq/sora-2/raw/main/cat.png" # Example image
)
prompt_input_image = gr.Textbox(
label="Enter your text prompt for the video",
placeholder="Describe the action or style you want for the video (e.g., 'The cat starts to dance')",
lines=3
)
generate_btn_image = gr.Button("πŸ–ΌοΈ Generate Video from Image", variant="primary", size="lg")
with gr.Column(scale=1):
video_output_image = gr.Video(
label="Generated Video",
height=400,
interactive=False,
show_download_button=True
)
status_output_image = gr.Textbox(
label="Status",
interactive=False,
visible=True
)
gr.Examples(
examples=[
["https://huggingface.co/spaces/akhaliq/sora-2/raw/main/cat.png", "The cat starts to dance"],
["https://huggingface.co/spaces/akhaliq/sora-2/raw/main/forest.png", "A magical forest where trees shimmer with light"],
["https://huggingface.co/spaces/akhaliq/sora-2/raw/main/car.png", "A classic car driving through a futuristic city"]
],
inputs=[image_input, prompt_input_image],
outputs=video_output_image,
fn=simple_generate_image_to_video,
cache_examples=False,
api_name=False,
show_api=False,
)
generate_btn_image.click(
fn=generate_image_to_video_with_pro_auth,
inputs=[image_input, prompt_input_image], # OAuth token is auto-injected by type hint
outputs=[video_output_image, status_output_image],
queue=False,
api_name=False,
show_api=False,
)
# Footer
gr.HTML("""
<div style="text-align: center; margin-top: 40px; padding: 20px; border-top: 1px solid #e0e0e0;">
<h3 style="color: #667eea;">Thank you for being a PRO user! πŸ€—</h3>
</div>
""")
def control_access(profile: Optional[gr.OAuthProfile] = None, oauth_token: Optional[gr.OAuthToken] = None):
"""Control interface visibility based on PRO status.
Gradio automatically injects gr.OAuthProfile and gr.OAuthToken based on type hints
when OAuth is enabled for the Space."""
if not profile:
# User not logged in
return gr.update(visible=False), gr.update(visible=False)
if verify_pro_status(oauth_token):
# User is PRO - show main interface
return gr.update(visible=True), gr.update(visible=False)
else:
# User is not PRO - show upgrade message
message = """
## ✨ Exclusive Access for PRO Users
Thank you for your interest in the Sora-2 Text & Image-to-Video Generator!
This advanced AI video generation tool is available exclusively for Hugging Face **PRO** members.
### What you get with PRO:
- βœ… Unlimited access to Sora-2 video generation (Text-to-Video & Image-to-Video)
- βœ… High-quality video outputs up to 1280x720
- βœ… Fast generation times with priority queue
- βœ… Access to other exclusive PRO Spaces
- βœ… Support the development of cutting-edge AI tools
### Ready to create amazing videos?
<div style="text-align: center; margin: 30px 0;">
<a href="http://huggingface.co/subscribe/pro?source=sora2_video" target="_blank" style="
display: inline-block;
background: linear-gradient(135deg, #667eea 0%, #764ba2 100%);
color: white;
padding: 12px 30px;
border-radius: 25px;
text-decoration: none;
font-weight: bold;
font-size: 1.1em;
box-shadow: 0 4px 15px rgba(102, 126, 234, 0.3);
transition: transform 0.2s;
">
πŸš€ Become a PRO Today!
</a>
</div>
<p style="text-align: center; color: #666; margin-top: 20px;">
Join thousands of creators who are already using PRO tools to bring their ideas to life.
</p>
"""
return gr.update(visible=False), gr.update(visible=True, value=message)
# Check access on load
# No explicit inputs are needed here as gr.OAuthProfile and gr.OAuthToken are
# provided automatically by Gradio to the function based on type hints.
demo.load(
control_access,
inputs=None, # Removed explicit instantiation of OAuthProfile and OAuthToken
outputs=[main_interface, pro_message]
)
return demo
# Launch the application
if __name__ == "__main__":
# Clean up any leftover files on startup
try:
cleanup_temp_files()
# Also try to clear Gradio's cache
if os.path.exists("gradio_cached_examples"):
shutil.rmtree("gradio_cached_examples", ignore_errors=True)
except Exception as e:
print(f"Initial cleanup error: {e}")
app = create_ui()
# Launch without special auth parameters and no queue
# OAuth is enabled via Space metadata (hf_oauth: true in README.md)
app.launch(
show_api=False,
enable_monitoring=False,
quiet=True,
max_threads=10, # Limit threads to prevent resource exhaustion
)