Output reference

FrameDigest outputs

FrameDigest creates two complementary outputs from the same locally processed video: a compact AI Handoff for visual review and a chronological Full Evidence archive for deeper inspection.

3 min max32 MiB max1 FPS Full EvidenceLocal processing

Create a screenshot pack

Compare outputs

Start with the job, not the file tree

AI Handoff

Best for: Attaching manageable visual context to ChatGPT, Claude, or another vision-capable workflow.
Optimizes for: Representative visual coverage with less automatic redundancy.
Includes: context.md, compact timestamped visual-sheet-XX.jpg files, every manually selected detail JPEG, and timeline.json inside the ZIP.

Manual AI Handoff selection is not capped by FrameDigest and selected moments are not silently dropped.

Full Evidence

Best for: Predictable chronological visual coverage and a broader source reference.
Optimizes for: Traceable one-frame-per-second timeline evidence.
Includes: Timestamped JPEG frames, contact sheets, frames.csv, context.md, and timeline.json.

At the current 180-second limit, Full Evidence preserves up to about 180 frames at fixed 1 FPS.

Decision guide

Which output should I use?

You need…	Use
A manageable set of visuals for an AI tool	AI Handoff
Every one-second visual checkpoint	Full Evidence
A quick chronological visual overview	Contact sheets
A manually chosen key moment	AI Handoff detail frames
Chronological filenames and CSV evidence	Full Evidence

Package anatomy

AI Handoff

<video>-ai-handoff/
├── context.md
├── visual-sheet-01.jpg
├── visual-sheet-XX.jpg
├── selected-<timestamp>.jpg
└── timeline.json

The compact package is designed to keep automatic visual coverage manageable while preserving every moment you selected manually.

Package anatomy

Full Evidence

<video>-full-evidence/
├── frames/
│   ├── frame_0001_00-00-00-000.jpg
│   └── ...
├── contact-sheets/
│   ├── sheet_001.jpg
│   └── ...
├── frames.csv
├── context.md
└── timeline.json

The chronological archive remains separate from the compact handoff so reduced redundancy never becomes silent evidence loss.

Reference

What the files are for

File / directory	Output	Purpose
`frames/`	Full Evidence	Chronological 1 FPS JPEG frames
`contact-sheets/`	Full Evidence	Multi-frame timeline overviews
`frames.csv`	Full Evidence	Frame index, timestamp, and filename mapping
`context.md`	Both where the contract specifies	Readable context and provenance
`timeline.json`	Packaged outputs	Structured chronology and visual-state provenance
`visual-sheet-XX.jpg`	AI Handoff	Compact automatic visual coverage
`selected detail JPEGs`	AI Handoff	One image for every moment selected manually

Time and speech

Timestamps stay explicit; transcripts do not get invented

Frame filenames keep timestamp-aware chronological order, including values such as 00-00-00-000.
AI-facing detail images can carry the approved MM:SS.mmm timecode footer while Full Evidence remains pixel-pristine.
No transcript.vtt or transcript.txt is included while production transcription is deferred.
The AI Handoff ZIP is convenience packaging. Unzip it before attaching the loose handoff files to an AI tool.

FAQ

Output questions

Which FrameDigest output should I use with ChatGPT or Claude?

Start with the AI Handoff when you want a compact set of visual files. Use Full Evidence when you need broader chronological coverage or source-level visual reference.

What is Full Evidence?

Full Evidence is the chronological 1 FPS archive containing timestamped JPEG frames, contact sheets, frames.csv, context.md, and timeline.json.

What is included in the AI Handoff?

The current handoff includes context.md, compact timestamped visual sheets, every manually selected detail JPEG, and timeline.json inside the ZIP.

Are manually selected moments ever removed?

FrameDigest does not cap manual selections or silently remove moments you explicitly chose. External AI tools may impose their own attachment limits.

Why are there no transcript files?

Production transcription is deferred. FrameDigest does not invent transcript.vtt or transcript.txt when no approved transcription path has run.

What do the timestamps mean?

Timestamps tie each visual back to its position in the source video so chronological evidence stays understandable after extraction.

Ready when you are

Turn the source into something you can inspect.

Create a screenshot pack