New
Marlin 2B
Turn video into scene descriptions and timestamped events. Marlin-2B helps you organize footage, build event timelines and add visual context to video libraries.
- Free tier available
- No subscriptions
- No credit card required
Capabilities
What you can do with Marlin 2B.
Technical Specifications
Max Video Duration
7200 seconds
How to use in deAPI
Integrate Marlin 2B into your app with a single API call.
Need an API key?
Sign up for free and get $5 in credits to start.
Frequently Asked Questions
Marlin-2B is a video-language model developed by NemoStation and based on Qwen3.5-2B. It describes visible scenes and actions, and associates events with time ranges.
Caption mode returns a scene description and a list of events. Each event has a start time, an end time and a short description. These outputs can become video metadata, an event timeline or input for a search index.
Marlin analyzes video frames. It does not transcribe the audio track. Here, captioning means describing visual content. Use a speech-to-text model such as Whisper Large V3 CT2 alongside Marlin when you also need a transcript or spoken subtitles.
Yes. Its visual analysis does not depend on the audio track. It can describe visible activity in silent clips, although anything that is only audible will be missing from the result.
Treat them as estimates. The model analyzes sampled frames, and event boundaries can drift or extend beyond the clip. Review the matching video segment and validate time ranges before using them for automatic edits.
- No subscription
- No credit card required
- Pay only for what you use
Try Marlin 2B — $5 for free
Get $5 in free credits and start generating with Marlin 2B.