How it works
Process the video locally, download the AI Handoff, unzip the package, attach the visual files to a vision-capable AI tool, and ask the question you care about. FrameDigest prepares the evidence but does not send it for you.
- 01Digest
Process the source locally and review the representative visual moments.
- 02Download
Export the compact AI Handoff while keeping Full Evidence available separately.
- 03Unzip
The ZIP is convenience packaging; extract the loose handoff files first.
- 04Attach
Add the visual sheets, manual detail frames, and context files to your chosen vision-capable tool, subject to that tool's own limits.
- 05Ask
Use the timestamped visuals as source material for the question or inspection you want the external AI to perform.
Why AI Handoff
A normal video can contain many repetitive frames. AI Handoff keeps a smaller set of representative visuals while Full Evidence remains available separately when you need broader chronological coverage.
A normal video can contain hundreds or thousands of source frames. Sending every source frame can overwhelm an image-based workflow with redundant material.
FrameDigest keeps predictable 1 FPS Full Evidence while its compact handoff independently scans for representative visual changes and preserves every moment you select manually.
The handoff is source material, not an AI answer. The downstream tool decides what the visuals mean.
What you get
The handoff contains timestamped visual sheets, manually selected detail frames, and context files that help another tool understand chronology without pretending the export itself is an AI answer.
- Timestamped compact visual sheets for automatic coverage.
- One detail JPEG for every moment selected manually.
- context.md and timeline.json for chronology and provenance inside the packaged handoff.
- Separate Full Evidence when you need the broader one-frame-per-second visual trail.
Privacy and control
FrameDigest stops at the handoff boundary. It does not call ChatGPT, Claude, Gemini, or another AI API, does not fabricate transcripts, and does not silently upload your selected media to an AI provider.
- FrameDigest does not currently transcribe audio or invent transcript files.
- The current visual production envelope is 180 seconds, 32 MiB, and up to about 180 Full Evidence frames at fixed 1 FPS.
- Manual AI Handoff selection is not capped by FrameDigest; external AI tools may impose their own attachment limits.
- Normal hosting-level request metadata can exist when the site loads, but application code does not intentionally send selected video or generated media to an AI provider or analytics service.
AI Handoff FAQ
Answers about what the package includes, why it is compact, how it relates to Full Evidence, and what happens when you attach the files to an external AI tool.
Does FrameDigest connect directly to ChatGPT or Claude?
No. FrameDigest prepares files locally and you choose whether to attach them to a separate AI tool yourself.
Why not send every frame to an AI tool?
A dense source can contain large amounts of redundant imagery. The AI Handoff reduces redundant automatic visual states while Full Evidence stays available separately.
Does the AI Handoff include audio or a transcript?
No transcript is currently produced because production transcription is deferred. The current handoff is visual source evidence plus chronology/context files.
Does using FrameDigest guarantee an AI will understand the video?
No. FrameDigest prepares timestamped visual evidence; interpretation quality and attachment limits belong to the external AI tool you choose.