Spaces:
Sleeping
Sleeping
Fix: Correct and simplify Gradio frontend for MCP integration
Browse files- backend/app.py +187 -0
backend/app.py
ADDED
@@ -0,0 +1,187 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import gradio as gr
|
2 |
+
import os
|
3 |
+
import httpx
|
4 |
+
from typing import Dict, Any
|
5 |
+
|
6 |
+
# --- Backend Client Functions ---
|
7 |
+
# These functions call the Modal/backend endpoints.
|
8 |
+
|
9 |
+
async def call_video_analysis_backend(video_url: str) -> Dict[str, Any]:
|
10 |
+
"""Calls the backend to analyze a single video."""
|
11 |
+
# Default to a placeholder if the env var is not set, to avoid crashing.
|
12 |
+
backend_url = os.getenv("BACKEND_VIDEO_URL", "https://your-backend-hf-space-for-video/process_video_analysis")
|
13 |
+
if not video_url:
|
14 |
+
return {"status": "error", "message": "Video URL cannot be empty."}
|
15 |
+
|
16 |
+
print(f"Sending request to backend for video: {video_url}")
|
17 |
+
payload = {"video_url": video_url}
|
18 |
+
try:
|
19 |
+
async with httpx.AsyncClient(timeout=1800.0) as client:
|
20 |
+
response = await client.post(backend_url, json=payload)
|
21 |
+
response.raise_for_status()
|
22 |
+
return response.json()
|
23 |
+
except httpx.HTTPStatusError as e:
|
24 |
+
return {"status": "error", "message": f"Backend Error: {e.response.status_code}", "details": e.response.text}
|
25 |
+
except Exception as e:
|
26 |
+
return {"status": "error", "message": "Failed to connect to backend", "details": str(e)}
|
27 |
+
|
28 |
+
async def call_topic_analysis_backend(topic: str, max_videos: int) -> Dict[str, Any]:
|
29 |
+
"""Calls the backend to analyze videos for a topic."""
|
30 |
+
backend_url = os.getenv("BACKEND_TOPIC_URL", "https://your-backend-hf-space-for-topic/analyze_topic")
|
31 |
+
if not topic:
|
32 |
+
return {"status": "error", "message": "Topic cannot be empty."}
|
33 |
+
|
34 |
+
print(f"Sending request to backend for topic: {topic} ({max_videos} videos)")
|
35 |
+
payload = {"topic": topic, "max_videos": max_videos}
|
36 |
+
try:
|
37 |
+
async with httpx.AsyncClient(timeout=3600.0) as client:
|
38 |
+
response = await client.post(backend_url, json=payload)
|
39 |
+
response.raise_for_status()
|
40 |
+
return response.json()
|
41 |
+
except httpx.HTTPStatusError as e:
|
42 |
+
return {"status": "error", "message": f"Backend Error: {e.response.status_code}", "details": e.response.text}
|
43 |
+
except Exception as e:
|
44 |
+
return {"status": "error", "message": "Failed to connect to backend", "details": str(e)}
|
45 |
+
|
46 |
+
# --- Gradio Tool Functions (Wrappers for MCP) ---
|
47 |
+
|
48 |
+
async def analyze_video(video_url: str):
|
49 |
+
"""
|
50 |
+
Triggers a comprehensive analysis of a single video from a URL.
|
51 |
+
|
52 |
+
This tool calls a backend service to perform multiple analyses:
|
53 |
+
- Transcribes audio to text.
|
54 |
+
- Generates a descriptive caption for the video content.
|
55 |
+
- Recognizes main actions in the video.
|
56 |
+
- Detects objects in keyframes.
|
57 |
+
|
58 |
+
:param video_url: The public URL of the video to be processed (e.g., a YouTube link).
|
59 |
+
:return: A JSON object containing the full analysis results from the backend.
|
60 |
+
"""
|
61 |
+
status_update = f"Analyzing video: {video_url}..."
|
62 |
+
results = await call_video_analysis_backend(video_url)
|
63 |
+
if isinstance(results, dict) and results.get("analysis") is None:
|
64 |
+
status_update = f"Error analyzing video: {results.get('error', 'Unknown error')}"
|
65 |
+
else:
|
66 |
+
status_update = "Video analysis complete."
|
67 |
+
return status_update, results
|
68 |
+
|
69 |
+
async def analyze_topic(topic: str, max_videos: int):
|
70 |
+
"""
|
71 |
+
Finds and analyzes multiple videos based on a given topic.
|
72 |
+
|
73 |
+
This tool calls a backend service that searches for videos related to the topic,
|
74 |
+
then runs a comprehensive analysis on each video found.
|
75 |
+
|
76 |
+
:param topic: The topic to search for (e.g., 'latest AI advancements').
|
77 |
+
:param max_videos: The maximum number of videos to find and analyze (1-5).
|
78 |
+
:return: A JSON object with the aggregated analysis results for all videos.
|
79 |
+
"""
|
80 |
+
status_update = f"Analyzing topic '{topic}' with {max_videos} videos... this can take a very long time."
|
81 |
+
results = await call_topic_analysis_backend(topic, max_videos)
|
82 |
+
if isinstance(results, dict) and results.get("results") is None:
|
83 |
+
status_update = f"Error analyzing topic: {results.get('error', 'Unknown error')}"
|
84 |
+
else:
|
85 |
+
status_update = "Topic analysis complete."
|
86 |
+
return status_update, results
|
87 |
+
|
88 |
+
# --- Gradio UI ---
|
89 |
+
|
90 |
+
with gr.Blocks(theme=gr.themes.Soft()) as demo:
|
91 |
+
gr.Markdown("# LLM Video Interpretation MCP")
|
92 |
+
gr.Markdown("This Hugging Face Space provides tools for processing video context for AI agents. Use the tools below to analyze videos by URL or by topic.")
|
93 |
+
|
94 |
+
with gr.Tab("Single Video Analysis"):
|
95 |
+
gr.Markdown("## Analyze a single video from a URL")
|
96 |
+
with gr.Row():
|
97 |
+
video_url_input = gr.Textbox(label="Video URL", placeholder="Enter a YouTube or direct video URL...", scale=4)
|
98 |
+
submit_button = gr.Button("Analyze Video", variant="primary")
|
99 |
+
status_text = gr.Textbox(label="Status", interactive=False)
|
100 |
+
json_output = gr.JSON(label="Analysis Results")
|
101 |
+
|
102 |
+
submit_button.click(
|
103 |
+
analyze_video,
|
104 |
+
inputs=[video_url_input],
|
105 |
+
outputs=[status_text, json_output],
|
106 |
+
api_name="analyze_video"
|
107 |
+
)
|
108 |
+
gr.Examples(
|
109 |
+
examples=["https://www.youtube.com/watch?v=3wLg_t_H2Xw", "https://www.youtube.com/watch?v=h42dDpgE7g8"],
|
110 |
+
inputs=video_url_input
|
111 |
+
)
|
112 |
+
|
113 |
+
with gr.Tab("Topic Video Analysis"):
|
114 |
+
gr.Markdown("## Analyze multiple videos based on a topic")
|
115 |
+
with gr.Row():
|
116 |
+
topic_input = gr.Textbox(label="Enter a topic", placeholder="e.g., 'Apple Vision Pro review'", scale=3)
|
117 |
+
max_videos_slider = gr.Slider(minimum=1, maximum=5, value=2, step=1, label="Number of Videos to Analyze")
|
118 |
+
topic_submit_button = gr.Button("Analyze Topic", variant="primary")
|
119 |
+
topic_status_text = gr.Textbox(label="Status", interactive=False)
|
120 |
+
topic_json_output = gr.JSON(label="Analysis Results")
|
121 |
+
|
122 |
+
topic_submit_button.click(
|
123 |
+
analyze_topic,
|
124 |
+
inputs=[topic_input, max_videos_slider],
|
125 |
+
outputs=[topic_status_text, topic_json_output],
|
126 |
+
api_name="analyze_topic"
|
127 |
+
)
|
128 |
+
gr.Examples(
|
129 |
+
examples=[["self-driving car technology", 2], ["open source large language models", 3]],
|
130 |
+
inputs=[topic_input, max_videos_slider]
|
131 |
+
)
|
132 |
+
|
133 |
+
# Set environment variables in your Hugging Face Space settings, not here.
|
134 |
+
# BACKEND_VIDEO_URL = "https://your-modal-or-backend-url/process_video_analysis"
|
135 |
+
# BACKEND_TOPIC_URL = "https://your-modal-or-backend-url/analyze_topic"
|
136 |
+
|
137 |
+
demo.launch() "https://www.youtube.com/watch?v=dQw4w9WgXcQ",
|
138 |
+
"https://sample-videos.com/zip/10/mp4/SampleVideo_1280x720_1mb.mp4"
|
139 |
+
],
|
140 |
+
inputs=input_text
|
141 |
+
)
|
142 |
+
gr.Markdown("**Processing can take several minutes** depending on video length and model inference times. The cache on the Modal backend will speed up repeated requests for the same video.")
|
143 |
+
|
144 |
+
with gr.Tab("Demo (for Manual Testing)"):
|
145 |
+
gr.Markdown("### Manually test video URLs or paths for interpretation and observe the JSON response.")
|
146 |
+
demo_interface.render()
|
147 |
+
|
148 |
+
with gr.Tab("Topic Video Analysis"):
|
149 |
+
gr.Markdown("### Analyze Multiple Videos Based on a Topic")
|
150 |
+
gr.Markdown("Enter a topic, and the system will search for relevant videos, analyze them, and provide an aggregated JSON output.")
|
151 |
+
|
152 |
+
with gr.Row():
|
153 |
+
topic_input = gr.Textbox(label="Enter Topic", placeholder="e.g., 'best cat videos', 'Python programming tutorials'", scale=3)
|
154 |
+
max_videos_input = gr.Number(label="Max Videos to Analyze", value=3, minimum=1, maximum=5, step=1, scale=1) # Max 5 for UI, backend might support more
|
155 |
+
|
156 |
+
topic_analysis_output = gr.JSON(label="Topic Analysis Results")
|
157 |
+
|
158 |
+
with gr.Row():
|
159 |
+
topic_submit_button = gr.Button("Analyze Topic Videos", variant="primary")
|
160 |
+
topic_clear_button = gr.Button("Clear")
|
161 |
+
|
162 |
+
topic_submit_button.click(
|
163 |
+
fn=call_topic_analysis_endpoint,
|
164 |
+
inputs=[topic_input, max_videos_input],
|
165 |
+
outputs=[topic_analysis_output]
|
166 |
+
)
|
167 |
+
|
168 |
+
def clear_topic_outputs():
|
169 |
+
return [None, 3, None] # topic_input, max_videos_input (reset to default), topic_analysis_output
|
170 |
+
topic_clear_button.click(fn=clear_topic_outputs, inputs=[], outputs=[topic_input, max_videos_input, topic_analysis_output])
|
171 |
+
|
172 |
+
gr.Examples(
|
173 |
+
examples=[
|
174 |
+
["AI in healthcare", 2],
|
175 |
+
["sustainable energy solutions", 3],
|
176 |
+
["how to make sourdough bread", 1]
|
177 |
+
],
|
178 |
+
inputs=[topic_input, max_videos_input],
|
179 |
+
outputs=topic_analysis_output,
|
180 |
+
fn=call_topic_analysis_endpoint,
|
181 |
+
cache_examples=False
|
182 |
+
)
|
183 |
+
gr.Markdown("**Note:** This process involves searching for videos and then analyzing each one. It can take a significant amount of time, especially for multiple videos. The backend has a long timeout, but please be patient.")
|
184 |
+
|
185 |
+
# Launch the Gradio application
|
186 |
+
if __name__ == "__main__":
|
187 |
+
app.launch(debug=True, server_name="0.0.0.0")
|