Documentary Montage
Retrieval-first thematic montage pipeline. Builds a semantic corpus of real-world footage from Pexels, Archive.org (Prelinger et al.), NASA, Wikimedia Commons, and Unsplash, then uses CLIP-based retrieval to fill slot descriptions from a thematic brief. The edit arranges clips by narrative beat with music sync and uniform color grade across mixed-era footage. Inspired by Adam Curtis / Chris Marker / Errol Morris tone poems.
- Stages
- 5
- Human approval gates
- 4
- Default budget
- $1.00
Stages
The agent reads each stage's director skill before working, then writes a checkpoint. Stages marked pause for your approval before the production continues.
- 1
Idea
Human approvalopenmontage/skills/pipelines/documentary-montage/idea-director.md
Produces
- brief
Tools
None — agent reasoning only
Review focus
- Thematic question is ONE sentence
- Tone register is ONE value from the fixed list
- Duration and shape are concrete
- Music plan is present (MANDATORY — silent only if user explicitly opted out)
- End-tag plan is present (MANDATORY — one philosophical closing line, rendered as Remotion end-card, default mode "overlay" on final scenes, unless user explicitly opted out)
- Narration plan is present (narration itself is OPTIONAL — absence is fine if music + visuals + end-tag carry the register)
Success criteria
- Schema-valid brief artifact
- thematic_question present in metadata
- music_plan present in metadata (source may be `none` ONLY with explicit user opt-out note)
- end_tag_plan present in metadata (text, palette, duration, mode — may be `null` ONLY with explicit user opt-out note)
- 2
Scene Plan
Human approvalopenmontage/skills/pipelines/documentary-montage/scene-director.md
Produces
- scene_plan
Tools
None — agent reasoning only
Review focus
- Slot descriptions use concrete noun-and-adjective language
- Every slot has 2-3 short search queries
- At least 2 slots are marked hero
- Slot count and target holds match the tone/duration math
- era_mix is reflected in preferred_sources distribution
Success criteria
- Schema-valid scene_plan artifact
- metadata.slots[] present with per-slot queries and preferred_sources
- Sum of target_hold_seconds within 10% of brief.duration_seconds
- 3
Assets
Human approvalopenmontage/skills/pipelines/documentary-montage/asset-director.md
Produces
- asset_manifest
Tools
- direct_clip_search
- corpus_builder
- clip_search
- music_gen
Review focus
- Every slot has exactly one picked clip
- No clip_id is picked for two slots
- Provenance (provider, original_url, license) present on every asset
- Standard path: corpus size >= 8x slot count, scores >= 0.22, diversify ran clean
- Fast path: thumbnails inspected, cross-act reuse noted, gaps filled
- rejected_picks log has reasons for passes (standard path)
Success criteria
- Schema-valid asset_manifest artifact
- One video asset per scene slot
- metadata.corpus_stats present (standard path) OR metadata.search_stats present (fast path)
- 4
Edit
Human approvalopenmontage/skills/pipelines/documentary-montage/edit-director.md
Produces
- edit_decisions
Tools
None — agent reasoning only
Review focus
- renderer_family is locked to documentary-montage
- Hero slots hold longest; mid-sequence cutaways shortest
- No two adjacent cuts share subject AND scale
- Transition vocabulary is at most 4 distinct values
- Music config is present (MANDATORY — silent only if brief.metadata.music_plan.source=none with explicit opt-out note)
- End-tag timing is present (MANDATORY — overlay mode: end_tag.offset_seconds in edit_decisions; concat mode: end-tag appended after body. Unless brief.metadata.end_tag_plan is null with explicit opt-out note)
- total_duration_seconds matches sum of cut durations (body only — end-tag overlay does not extend duration; concat mode adds tag duration)
- Every cut has a reason
Success criteria
- Schema-valid edit_decisions artifact
- Timeline duration within 10% of brief.duration_seconds
- 5
Compose
Auto checkpointopenmontage/skills/pipelines/documentary-montage/compose-director.md
Produces
- render_report
Tools
- video_compose
- audio_mixer
- color_grade
- video_trimmer
- video_stitch
Review focus
- Output duration matches planned within 1s (body duration for overlay mode; body + tag for concat mode)
- Resolution matches target_platform canvas
- Uniform LUT applied across the timeline
- Music is mixed in (MANDATORY — silent output only with explicit user opt-out recorded in brief)
- render_runtime locked at proposal MUST stay 'remotion' on this pipeline until HyperFrames end-tag parity lands — the overlay/concat end-tag stack depends on Remotion CinematicRenderer components. Silent swap to hyperframes or ffmpeg is a CRITICAL governance violation.
- End-tag rendered via Remotion (MANDATORY — absence only with explicit opt-out). Default mode: overlay (ProRes 4444 with alpha composited on final scenes). Fallback: concat (opaque card appended after body).
- Overlay mode: extract a frame from the overlay region and verify text is visible over footage, not over black. Concat mode: last frame must be the end-tag card.
- No silent fallback from a motion-led promise
Success criteria
- Schema-valid render_report artifact
- Output file exists and passes ffprobe validation
- render_report.end_tag_rendered = true (or explicit opt-out)
- render_report.end_tag_mode = "overlay" or "concat"
- render_report.music_mixed = true (or explicit opt-out)
Manifest: openmontage/pipeline_defs/documentary-montage.yaml. Director skills live under openmontage/skills/.