JoyAI VL Live
Disconnected
Dark
Nhật ký gỡ lỗi
Sao chép
Xóa
Đóng
Chưa có log.
Settings
×
Layout
Main Content Order
Choose which element appears at the top
Camera → VLM Output Info
VLM Output Info → Camera
VLM Output on Camera View
Show text overlay directly on video feed
None
At the top
At the bottom
Visual Effects
Pop-in Animation
Scale animation when new VLM response arrives
Green Glow Effect
Border glow on new VLM response
Fade Effect
Gradually fade response after 2 seconds
Visual Style
Colorful UI Accents
Color-coded icons and input focus glows
WebRTC
Max Video Latency (seconds)
Drop old frames if delay exceeds this (0 = no intervention)
Audio Output
Speak VLM output
Play TTS audio for each visible response
Background Model
Enable delegation solver
Run Qwen3.5-122B-A10B-FP8 for delegated questions, visual reasoning, and chart tasks in the background
Frame multiplier
Background frames per second relative to foreground streaming FPS
Max background frames
Recent background frame cache cap; default and maximum are 100
Debug
Show request payload
Include request JSON (image + prompt) under the prompt area; collapsed by default
Show response payload
Include API response JSON under the VLM output; collapsed by default
Show memory state
Display mid-term and long-term memory content below VLM output
VLM API Configuration
▼
API Base URL
Ollama
http://localhost:11434/v1
vLLM
http://localhost:8000/v1
SGLang
http://localhost:30000/v1
OpenAI
https://api.openai.com/v1
NVIDIA API Catalog
https://integrate.api.nvidia.com/v1
Current VLM endpoint
API Key (Optional)
▼
Required for OpenAI and NVIDIA API Catalog, etc.
Model Selection
Loading models...
Video Source
▼
Webcam
RTSP Stream
BETA
Camera Selection
Detecting cameras...
Select camera device to use for VLM analysis
RTSP Stream URL
ℹ
Format:
rtsp://[user:pass@]ip:port/path
Examples:
• rtsp://192.168.1.100:554/stream
• rtsp://admin:password@192.168.1.100:554/h264Preview_01_main
Beta:
Tested with Reolink RLC-811A. Other cameras may work.
Learn more
Auto Analyze
Analyze continuously and report notable changes
Processing Interval
Seconds between each VLM inference (default: 1s)
Frames per Batch
Number of frames batched per inference (default: 1)
Start
后置
VLM Output
Ready
Mirror
VLM Output Info
Model:
--
Speaking:
Latency:
--
ms
Avg:
--
ms
Count:
--
Request payload (debug)
Markdown
Response payload (debug)
Mid-term memory
Long-term memory
Mô tả biểu cảm khuôn mặt và cử động của tôi theo thời gian thực
Theo dõi trực tiếp
✓
Tạo một hình ảnh phong cách màu nước dựa trên nội dung trong khung hình
Tạo ảnh theo cảnh
✓
Hãy chơi trò hỏi đáp nhanh trong 30 giây; trả lời thật nhanh và nhắc tôi khi hết giờ
Trò chơi 30 giây
✓
Mỗi khi xuất hiện một chai, hãy mô tả hình dáng của nó
Nhận diện vật thể
✓
按住说话