BVS is the product: a studio service that takes any client brief → finished video. ChapterCut was a single-project experiment — we harvest its proven 8-role crew and generalize it (chapter→shorts becomes brief→any video), then layer in the strengths BVS already has that ChapterCut never built. This is the crew the service runs on every client job.
| Role | Persona | Does | Status |
|---|---|---|---|
| 1 · Script Supervisor | — | Chapter/brief → beats, character roster, locations, hooks | ChapterCut |
| 2 · Casting Director | Sol | Locked casting sheets + 5-expression libraries + Higgsfield Soul IDs (character continuity) | ChapterCut |
| 3 · Production Designer | — | Locked sets, props, vehicles + world/era rules (text bibles) | ChapterCut + Bezar 3D upgrade |
| 4+5 · DP + Director | Jack | Shot plans (camera/lens/light) + per-character direction; scene_geography | ChapterCut |
| 6 · Prompt Engine | ATLAS | Pure-JS assembler → image/motion/negative prompts (injects bibles verbatim) | ChapterCut |
| 7 · Model Selector | Priya | Routes shot type → best image model (spec only, not yet wired) | spec only |
| 8 · Captionist | Tomo | Word-timed SRT + karaoke ASS, platform cuts | ChapterCut |
scene_geography + hardcoded "left seat/right seat" prefixes — a known weak spot). The Blender greybox + camera blocking + depth-ControlNet fills exactly that gap: it gives Production Designer a real geometric source of truth so space and props don't drift between shots.

Your Set Designer (where) + Prop Master (what) split lives inside Role 3: the set-spec pins geometry, a prop-bible pins each hero object's look — both fed to prompt assembly so the fridge/truck/gun are the same object every shot. Character continuity (your "Cast & Wardrobe") is already Role 2 (Casting Director + Soul IDs).
| Capability | ChapterCut status | Bezar has |
|---|---|---|
| 3D spatial / previs (greybox) | missing | /set-designer + greybox_kit |
| Motion QA (2-gate: frame + video-native) | none | motion_report + gemini_video |
| Sound design & music bed | not started (gate 0) | SFX + music (brand sting) |
| Model routing (Role 7) | spec only | bake-off data (Kontext>depth>gemini>gpt) |
| Brand identity (sting/favicons) | n/a | _brand_motion / _brand_assets |
The same crew handles any video — genre just changes which roles lead and which production track runs. The roles are constant; the playbook adapts.
| Track | For | Production | Roles that lead | Status |
|---|---|---|---|---|
| A · Instructional / technical | Polyguard install videos, explainers | Blender toon illustration + labels/overlays + VO | Production Designer (spec-accurate parts) · Motion QA · Captions/labels · VO | most proven (Insulrap shipped) |
| B · Narrative / photoreal | Brand spots (Big Gorilla), story adaptations | AI greybox → dress → i2v | Casting Director · DP+Director · Set/Prop (3D) · Shot Dresser | proving out now |
Build the crew as generalized, invokable skills in this repo (.claude/skills/), harvesting ChapterCut's role prompts as the starting point: (1) port + generalize Roles 1–6 + 8 (brief→video, client-agnostic); (2) fold the 3D greybox in as Role 3's spatial layer (already built: /set-designer); (3) make Motion QA and Sound/Music first-class roles (BVS-native, missing in ChapterCut); (4) wire Role 7 routing from the bake-off data; (5) add a Producer skill to chain the crew + client intake/delivery for the service. One product, one crew, run per client.