thativanng/reels-maker

voxel-walk: one person, one continuous walk, every role

Where the kit lives: run every command in this skill from the reels-maker folder (the clone of github.com/thativanng/reels-maker that holds template/ and tools/). If the project has none, set it up first by following that repo's SETUP.md.

Start from template/film.html + template/career.json. The template is a complete three-chapter film (a café, a warehouse, a night desk with an AI buddy) for a made-up person. Replace the chapters with the person's own; keep the systems.

1 · Interview first (reel-intake, then references/intake.md)

reel-intake runs the interview: purpose, format, the journey block (every role and hat, name/describe/hide, claims), the look references and the guardrails. This skill's references/intake.md adds only what the voxel style needs. The template is 4:5 only (its cameras are keyed for it): for a 9:16 brief, re-key the cameras and re-run every gate, or pick another look. Before a single block is placed, have from the person, in their words:

• Each role: the year, company, title, and the one thing that role taught or proved (the "moat").
• The scene that shows the work, not the job title. A planner isn't a person at a desk; it's trucks leaving a dock for the right depots. A barista is cups handed over a counter.
• The outfit for each chapter, what they actually wore.
• What must NOT appear: numbers they can't publish, an employer they'd rather not name, claims their current colleagues would question.
• The end card: name, tagline, three strengths, one closing line.

Their words outrank yours. You may proofread their lines (grammar, flow); you may not rewrite their meaning.

2 · The house rules (references/house-rules.md has the reasons)

1. One continuous walk. The hero never vanishes. They walk from scene to scene; the outfit changes mid-walk with a light ring sweeping head to toe.
2. Scenes rise and sink. The next chapter rises out of the floor ahead of the hero and the last one sinks behind, so a walk between chapters takes 2–4 s, not 20. Moving things rise after the buildings and leave before them.
3. The picture tells the story; the subtitle carries the meaning. No labels on objects (a forklift doesn't need a "FORKLIFT" tag). Label only a concept the picture can't say. Subtitles are one line of story per beat, never a repeat of the banner.
4. Screen time follows importance, not real scale. The hero's moments get holds; vehicles are quick supporting players.
5. Colour: warm grey ground, white objects, dark roofs and roads, ONE accent colour for the UI and the key props. Strong key light plus ambient occlusion. White on white is dull.
6. The banner changes when the scene is on screen, not when its code starts.
7. No number you can't defend in front of your current employer. Years on the banner are fine; KPIs usually aren't.
8. Every frame passes film-audit.

3 · How the template is built (find these sections in film.html)

• CHARACTERS: a character is a list of rects [x, y, w, h, colour, part, frontOffset, depth], extruded to boxes and grouped into parts (legs, torso, arms, head, hair) with pivots. career.json → looks gives one outfit per chapter. makeMini() makes the cast; buddyVox() makes the robot.
• The outfit wipe: each outfit is a separate voxel body clipped by a moving plane; WIPES = [[t, from, to]].
• WALK: IK = [[t, x, z, ryWhenStopped]] waypoints, trapezoid speed, leg phase from distance. SITS = [[from, to]] for desks.
• ZONES: zoneG(tIn, tOut) makes a chapter that rises and sinks; anything that moves goes in wrap(zone).
• CAMERA: CK = [[t, A(...) | F(...)]].
• OVERLAYS: all text comes from career.json.
• AUDIT: AUD_S (solids), AUD_M (movers), HELD (contacts that are meant to happen), window._dbg (what probes can measure).

4 · Adding a chapter

1. Pick a strip of street (x range) next to the last chapter. The walking lane is z ≈ 2.6–3.4, props stand behind it (z < 0), a desk sits in front (z ≈ 5.4) with its chair behind.
2. Build the set in a zoneG(tIn, tOut); put movers in wrap(zone).
3. Add IK waypoints for the walk in and the stop, a WIPES entry for the outfit, and CK keys (follow while walking, an absolute hold on the beat).
4. Add the era to career.json at the moment the set is up and the outfit is on. Add the subtitles.
5. Register every solid in AUD_S and every mover in AUD_M. Add the audio cues in template/audio.py.
6. Run the gates (stills sheet → node tools/collide.mjs → probes), then an independent auditor.

5 · Common traps in this style

• A prop behind the head (shelves, screens, caps on a production line) reads as a hat or a halo. Probe it. Fix it with camera height, a sidestep, or a later move.
• Pairs of people standing one behind the other fuse into a two-headed figure. Space them a body-width apart in x.
• Things handed between people must stay at chest height and never cross a face. Route them through the hero's desk if they'd cross the room.
• The hero walking sideways sweeps their shoulders across things beside the lane. Keep 1.3 units of clearance.