READY
Frames are extracted in your browser. The video file itself is never sent to our server.
reading description, comments, frame…
Paste a YouTube link or upload a video — get the real prompt, or an honest AI guess.
HOW IT READS A VIDEO
Three passes, in this order — because order matters here.
01
Check the description
Many AI-video creators post their prompt right in the video description. We scan it first, because a disclosed prompt beats any guess.
02
Check top comments
If it’s not in the description, creators often drop it in a pinned or top comment. We check there next.
03
Reconstruct from the frame
If neither exists, an AI vision model studies the video’s thumbnail and writes its own best-effort prompt — clearly marked as a reconstruction, not fact.
TWO KINDS OF RESULT
We never blur the line between the two.
Questions
A tool that looks at an existing video and gives you a prompt that describes it — either the creator’s own disclosed prompt, or an AI’s best-effort reconstruction. This one works specifically with YouTube links and is explicit about which one you’re getting.
No tool can — YouTube doesn’t store generation prompts, and video pixels don’t encode the text that made them. We show you a disclosed prompt when the creator shared one, and an honestly-labeled reconstruction when they didn’t.
Yes — the tool doesn’t care which AI generator made the video, only that it’s hosted on YouTube. It reads the video’s own description, comments, and thumbnail, whatever tool made it.
Yes, free to use.
Yes. Frames are extracted directly in your browser and only those still images are sent for analysis — the video file itself is never uploaded to our server. Uploaded-video results are always labeled “Reconstructed,” since there’s no description or comments to check for a disclosed prompt.
No — deliberately. A still frame can’t actually reveal lens type, camera model, or lighting equipment, so we don’t let the AI guess at those and present it as a finding. Reconstructions stick to what’s genuinely visible: framing, subject, color, mood, and setting.
No. Because frames are extracted in your browser rather than uploaded to a server, there’s no server-side file size ceiling — 4K, 8K, or long videos work the same way a short clip does. Most competing tools cap uploads around 20–50MB; this one doesn’t need to, since it never receives the file at all.
Yes — every reconstructed result comes with three versions, phrased in the typical style of Sora, Kling, and Runway prompts. They all describe the same underlying guess; only the wording convention changes. We don’t claim one model would work “better” for your specific video — that’s not something a still image can actually tell you.
Not yet via link — but you can upload a TikTok or Instagram video file directly using the upload option, which works with any video file regardless of where it came from.