infery
← All models

Marlin Find

marlin-find

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?

Details

Accepts
video

Pricing

Input
1875 cr / 1M tokens

Prices in credits (1 credit = $0.01).

Data schema

Input

FieldTypeDescription
top_pnumberNucleus sampling threshold. Only used when do_sample is true.
promptrequiredstringNatural-language description of the event or scene to find in the video.
do_samplebooleanIf true, sample with temperature/top_p; if false, use greedy decoding.
video_urlrequiredstringURL of the video to search through.
temperaturenumberSampling temperature. Only used when do_sample is true.

Output

FieldTypeDescription
spanDetected (start, end) span in seconds, or null if the model could not locate the event.
textstringRaw post-thinking model output (e.g. "From 14.3 to 18.2.").
format_okbooleanTrue iff the output matched the trained "From X to Y." format.