Zero to Director #9 — The Full Reference Desk: Image, Audio & Video Conditions
From Last Time
Post #8 carved motion from words alone — and words are a weak instrument for time. "Slow push-in" says a lot about the camera and nothing about the scene you already have, the face you already cast, the soundtrack already playing in your head, or the _move you saw in someone else's film.
The machine from post #8 — H3, the ref2va carver — has a middle name that matters: reference-to-video-audio.
This post opens the reference desk.
Why References
Every arm of the desk is the same argument the bust made in post #5, moved onto the time axis: show, don't tell.
- The sentence "start from this exact frame" is impossible — the image is the instruction.
- "Keep this person on screen" is fragile in words — a face is the actor.
- "Cut to the beat" cannot be said at all — an audio clip is the timing.
- "Move like that shot" would take a paragraph — a video clip is the choreography.
The Three Arms
Beside the text blueprint, the desk takes up to three kinds of conditioning, each a station on its own rail into the carver:
- The image arm — a first frame to grow from, or an identity to hold. A photograph is the strongest scene description there is: composition, light, and subject locked before the first cut.
- The audio arm — sound that paces the carving. H3 casts synchronized sound natively; give the arm a clip and the cuts follow it — footsteps, drums, speech.
- The video arm — motion transfer: borrow the movement of a reference roll — the path of a camera, the swing of an arm — and re-dress it with your own scene.
Each arm carries a weight: how loudly it speaks against the blueprint. And arms stack — the full desk holds a first frame and a face and a beat at the same time.
Composing Conditions
The composition rules are the ones the series has been teaching all along:
- Complementary promises compose; competing promises argue. A first frame that sets the workshop and a beat that sets footfalls agree; two arms both trying to own the camera fight.
- Weakest arm first. When a composed run surprises you, drop the arm weights one at a time — the diagnostic habit from every dial so far.
- The blueprint becomes the narrator. With arms carrying scene, face, and timing, the words are free to direct: mood, intent, what happens next.
Examples
The first frame
09-first-frame.json grows a roll from a still you have already carved — the post-#3 heron photograph as frame one.
Composition locked before the first cut: {OBSERVE-FIRSTFRAME}
Identity on set
09-identity-on-set.json brings post #5's casting trick to video: one face reference, held across a moving shot.
The actor contract, now in time: {OBSERVE-IDENTITY}
Sound-led
09-sound-led.json hands the pacing to a clip: the arm takes audio, and the carve follows it.
What a beat does to a chisel: {OBSERVE-SOUND}
Motion transfer
09-motion-transfer.json borrows movement from a reference roll and re-dresses it: same choreography, new scene.
Structure from wire, look from bust — and now motion from film: {OBSERVE-MOTION}
The full desk
09-full-desk.json composes the arms: first frame, identity, and beat together, blueprint narrating.
The composition, measured: {OBSERVE-FULL}
Pitfalls
- Arms fight like dials do. Two arms owning the same dimension — two cameras, two timings — produce mush. Compose complementary promises only.
- The first frame will not freeze forever. A roll that starts at your frame drifts by design; if you need the exact still, keep it as a still. Want less drift — say so in the blueprint.
- Identity is a strength dial, not a law. Too low and the face wanders between frames; too high and every frame becomes a portrait session. Watch, adjust, re-run.
- Cheap audio in, cheap timing out. The arm hears what you feed it — noisy clips produce confused cuts.
Your Move
Grow a roll from your favorite still of the series, then add one arm at a time — face, then beat, then a borrowed move — and watch each one steer.
Next post: toolboxes again, but for the video machine — turbo boxes that buy back the one thing video costs most, time.