← All models
Marlin
marlin
Visionby marlin
Marlin is a 2B video VLM tuned for the two questions developers actually want to ask of their videos: what is happening, and when?
Pricing
- Input
- 1875 cr / 1M tokens
Prices in credits (1 credit = $0.01).
Data schema
Input
| Field | Type | Description |
|---|---|---|
| top_p | number | Nucleus sampling threshold. Only used when do_sample is true. |
| promptrequired | string | Caption prompt sent to the model. The example value is Marlin's canonical training prompt — overriding usually degrades output quality. |
| do_sample | boolean | If true, sample with temperature/top_p; if false, use greedy decoding. |
| video_urlrequired | string | URL of the video to caption. Up to ~2 minutes is supported. |
| max_tokens | integer | Maximum number of tokens to generate for the caption. |
| temperature | number | Sampling temperature. Only used when do_sample is true. |
Output
| Field | Type | Description |
|---|---|---|
| text | string | Full post-thinking caption text (Scene + Events) as returned by the model. |
| scene | string | Spatial description of the clip. |
| events | array | Time-ranged events parsed from the caption. |