skills-shared/reference-analysis.md

# Reference Information

Use available reference information to guide the requested creative work.

## Video Reference Information

For a supplied file or a link the user asks to work from, use `./reference-breakdown.md`:
probe, frames, word-labelled contact sheets, cut candidates, local transcription and
timecoded notes (`ANALYSIS.md`, `TIMELINE.md`). The remake workflow rebuilds the piece from
those notes; other workflows consume them for scripts, boards and adaptation.

Reuse available information about mood, pacing, camera choices, text, sound and the moments
the user wants to preserve or change. Keep observed details distinct from creative
assumptions; do not claim to have watched footage based on a link or summary alone.
A complete shot-by-shot report is not required to begin planning from sufficient reference
information. For exact source-preserving edits, obtain usable footage before editing.

## Image Reference

Classify it:

- identity anchor: preserve subject
- brand anchor: preserve logo/packaging/visual system
- style reference: preserve aesthetic only
- layout reference: preserve composition
- raw asset: use directly or upload

If the image is an identity/brand anchor, pass it to PixVerse as an image reference instead of only describing it.

## Storyboard Boards And Dossier Images

When a generated board contains panels, labels, arrows, column headers, frame numbers, borders, or split-screen layout, treat that board furniture as reference-only. The video prompt should say the model should read those elements for identity, sequence, composition, and motion intent, not reproduce the literal grid.

A multi-panel board usually means sequential beats over time. If several subjects must appear together, describe one cohesive frame or space; do not ask the video model to render a panel collage unless the user explicitly wants a split-screen graphic.