Qwen3-VL Tester

1 · Upload clips

📦

Drop a .zip here

or click to browse

Previous sessions

Past runs

2 · Model

Qwen3-VL 8B FP8 Qwen3-VL 4B FP8

3 · Extraction strategies

Full video

Entire clip via Gradio — model samples internally

Uniform N frames

Evenly spaced across the clip

Fixed FPS sampling

Extract at a target frames-per-second

Scene detection

Histogram diff at cut boundaries

Motion peaks

Optical-flow: highest-motion moments

Pixel diff keyframes

Frames where pixel diff > threshold

First / Mid / Last

3 frames: start, middle, end

Grid composite

N frames stitched into one grid image

4 · Parameters

Frames (N)

Target FPS

Start offset (s)

Max image dim (px)

Max output tokens

5 · Prompt

Running… 0%

Results 📅 Summary appears when processing completes

🎬

Upload a zip of clips, choose strategies, then hit Run analysis.

⚙ Jobs

No jobs yet.

✕

📅 Summary

Best with Structured JSON prompts · Alert+Detail gives alert text only

Loading…

⚠ Anomaly Scoring

Scoring clips…

🎬 Generated Videos

Loading…