Qwen3.6-35B-A3B on ZeroGPU · vs Amazon Nova 2 Lite
Raw footage in. Highlight reel out.
Pick a clip and a sampling rate. Both models get the same frames, and you get a playable shot list, the latency, the token usage, and the cost per video-hour at your settings. The video file stays in your browser; only the sampled frames are sent.
View llms.txt
Paste it into Claude Code, Cursor or ChatGPT to run this benchmark on your own clips. All it needs is a ZeroGPU API key.
01Footage
Drop a clip here
or · MP4 or MOV
or
02Sampling
Recommended1 fps · 512 px: the best balance of detail and tokens for a first test.
Choose a clip to see frame count and tokens.
Qwen3.6 on ZeroGPU · — tok/frame—per video-hour, input · $0.20 / $1.50 per 1M tokens
Nova 2 Lite · ≈240 tok/frame—per video-hour, input · $0.30 / $2.50 per 1M tokens
03Task
Choose a clip to start.
Unlocked for ·
Your original clip plays here. Each model’s highlight reel gets its own player below.
The whole integration
OpenAI-compatible: keep your client, change the base URL and the model name.