New: Coeus, our open Video Jev model. It watches and listens to a video and answers questions about it, with a probability you can trust. Meet Coeus.

Research preview - free API key

Ask a video anything. Get probabilities back.

Coeus, our open Video Jev model, watches and listens to a clip and answers your typed questions with calibrated probabilities. Sign up for a free key: 10 videos a day.

Coeus

What you can ask

Yes or no

Ask anything you can see or hear.

Is there a face in the first three seconds? Is this AI-generated? You get the probability, not a guess in prose.

Scores

Rate on your own scale.

Give the levels in words. Coeus returns a probability for each level, so you can see how sure it is.

Moments

Find when something happens.

Ask for the funniest moment or the product shot. You get a probability for each second.

One pass

Every question, one forward pass.

Up to 60 s of video with audio. Ask ten questions for the price of one.

Request

curl https://models.retina.dev/v1/score \
  -H "Authorization: Bearer $VJEV_KEY" \
  -F file=@clip.mp4 \
  -F 'questions={"state": {}, "questions": {
    "speech":  {"type": "noul",  "instructions": "Is someone talking in this clip?"},
    "ai":      {"type": "noul",  "instructions": "Is this clip AI-generated?"},
    "quality": {"type": "score", "instructions": "Rate how clean and sharp this video looks.",
                "criteria": ["bad", "average", "good", "perfect"]},
    "text":    {"type": "moment", "instructions": "When does text first appear on screen?"}}}'

# Or send a link instead of a file:
#   -H "Content-Type: application/json" \
#   -d '{"video_url": "https://.../clip.mp4", "questions": {...}}'

Response

{
  "answers": {
    "speech":  {"type": "noul", "answers": {"yes": 0.496, "no": 0.504}},
    "ai":      {"type": "noul", "answers": {"yes": 0.765, "no": 0.235}},
    "quality": {"type": "score",
                "answers": {"bad": 0.006, "average": 0.010, "good": 0.368, "perfect": 0.616},
                "expected": 3.594},
    "text":    {"type": "moment",
                "bins": [[0, 1, 0.035], [1, 2, 0.080], [2, 3, 0.116], ...],
                "span": {"start_s": 2.0, "end_s": 8.0, "peak_s": 7.0, "p_peak": 0.1779, "p_span": 0.8858}}
  },
  "seconds_read": 8.0,
  "usage": {"used_today": 1, "limit": 10}
}

Limits: 10 videos per key per day, the first 60 s of each video, files up to 50 MB.

Research preview for non-commercial use, under CC BY-NC-SA 4.0. Need more? Run the open model yourself from github.com/sp1d5r/vjev.