[P1][Creator Video] Edit uploaded footage into structured Reels/Shorts with VideoNarrative, VideoEditPlan, captions, and deterministic rendering

[Ankit6149]

Parent epic

#151

Product outcome

SignalFlow can turn user-uploaded raw footage into a reviewable short-form video without requiring a professional nonlinear editor for common content tasks.

Example:

Make a clean 25-second Instagram Reel from these clips. Start with clip 3, use the close-up from clip 1, remove dead space, keep transitions calm, and add subtitles.

Canonical docs

• docs/MEDIA_INTELLIGENCE_AND_CREATIVE_PRODUCTION.md
• docs/CREATIVE_MEDIA_DOMAIN_CONTRACTS.md

Pipeline

raw clips
→ immutable Asset ingestion + policy/rights
→ proxy/thumbnails where useful
→ transcription/audio analysis
→ shot/scene understanding
→ usable-moment selection
→ VideoNarrative
→ VideoEditPlan
→ deterministic render
→ captions/overlays/audio/cover
→ quality/privacy/rights review
→ exact revision approval

Basic creator-editing scope

Support common automation such as:

• trim;
• remove obvious dead space;
• reorder/join clips;
• crop/reframe;
• 16:9 → 9:16 adaptation;
• captions/subtitles;
• title cards/overlays;
• logo/brand treatment;
• simple cuts/transitions;
• simple speed changes;
• zoom/focus emphasis;
• basic audio leveling/ducking;
• user-provided/licensed music;
• intro/outro;
• cover/thumbnail;
• platform derivatives.

Structured contracts

Implement:

• VideoNarrative — story meaning/beats;
• VideoEditPlan — exact timeline/clip/source/crop/speed/overlay/caption state;
• CaptionTrack;
• AudioPolicy;
• immutable rendered MediaComposition revisions.

Natural-language instructions must mutate structured plans.

Examples:

"the first three seconds are boring"
→ revise opening timeline segment

"use clip 3 first"
→ reorder stable source bindings

"captions are too big"
→ style-only change

"transitions are too flashy"
→ transition-policy change

AI vs deterministic responsibility

AI may:

• understand footage;
• identify strong moments;
• plan narrative/sequence;
• write captions/overlays;
• recommend crop/focus.

Deterministic media tooling should perform:

• trim/cut/order;
• crop/reframe;
• captions;
• transitions;
• audio mix;
• overlay/brand rendering;
• encoding.

Generative video is not required for this issue.

Acceptance criteria

• Multiple uploaded source clips retain immutable lineage/rights/use policy.
• At least one real short-form narrative is planned from footage through structured records.
• VideoEditPlan references exact source versions and valid source in/out ranges.
• Natural-language revision updates only dependent plan state where possible.
• Captions/transcript timing is reviewable/correctable.
• At least one 9:16 final render is produced through deterministic timeline execution.
• User can request a duration/order/caption/transition change without rebuilding from scratch.
• Cancellation/retry/restart behavior is durable for long media jobs.
• Privacy/rights/identity-sensitive restrictions are enforced.
• Exact final video revision and cover are approval-bound.
• Original footage is never overwritten.

Non-goals

• no Premiere/DaVinci replacement;
• no advanced masking/VFX/node graphs;
• no full professional color suite;
• no DAW-level audio system;
• no unrestricted face/voice transformation;
• no generative-video dependency.

Definition of done

Close only when a real set of user clips can become a professional basic Reel/Short through a traceable structured edit plan, natural-language revision, deterministic render and exact approval without destroying source media or requiring manual timeline editing for the closing proof.