Film Method KB

📐 Film Method KB

The standing production-method rulebook — how to write voices, build location plates, lock the look, structure a video prompt, choose optics, hold scene geography, and direct acting. Editable and cascading; this is the memory the video-prompt tools consult. Grow it as you learn what to lock in.

Editing:

Voice & accent ○ Built-in default

How a character's voice is written — as a fixed block decided before the dialogue and pasted verbatim, never reworded.

↳ feeds: Voice profiles · CINEDANCE (AUDIO block)

Locations & the location kit ○ Built-in default

How a location plate is generated (three-quarter, anchored, one light) and how the full angle kit is built from one image.

↳ feeds: LIRA (location) · Location Lab / Set Map · CINEDANCE (LOCATION MAP)

LOCATIONS — generate a plate as a WIDE or MEDIUM in THREE-QUARTER view, never frontal. A frontal room is flat wallpaper the model cannot read volume from, and it reinvents everything past the frame edges; three-quarter gives depth, places the heroes correctly, and yields almost a full circle of angles from one plate.
Leave a visible ANCHOR in every location — a column, a lamp, a sofa, a crooked chair — and tie all staging to it ("the hero at the lamp, facing the door" works; "the hero in the room" is a lottery). Keep ONE light logic: one source, one direction of shadows, never two suns. NO people and NO weapons in the plate. Fight the render look with the language of real surfaces: rust, cracks, tape, fingerprints, oil stains, water marks.
Pull a reverse angle out of VIDEO: generate a people-free clip of the camera slowly walking through the space, screenshot the angle you need, and bring it into Seedream or Nano Banana to fix texture and light — a full location kit out of one image. A DIALOGUE location needs the kit — three-quarter, front, reverse, and a background plate behind each hero in the scene; a PASS-THROUGH location needs one angle.

Look & colour (the visual bible) ○ Built-in default

How the palette is decided in prep — worlds with one register each, the 80/15/5 split, accents found as real in-frame sources.

↳ feeds: Director & DP (look-lock) · LIRA · CINEDANCE (STYLE / LIGHTING)

The 15-block prompt skeleton ○ Built-in default

The fixed block order every serious video shot is written in — plus the no-negative rule and the two reference rules that break generations.

↳ feeds: CINEDANCE (the writer)

THE 15-BLOCK SKELETON — every serious shot is written in the same blocks, in the same order:
SCENE CONTEXT · ACTIVE REFERENCES · LOCATION MAP · FIRST FRAME AND SPATIAL BLOCKING · FORMAT MODE · OPTICS · CAMERA · ACTION TIMING · PHYSICS · LIGHTING · AUDIO · CHARACTER ACTING · STYLE · QUALITY · POSITIVE CONSTRAINTS.
There is NO negative block. Write every prohibition as the desired OUTCOME, because "does NOT fall on his back" reads to the model as an invitation to think about falling on his back — write "falls on his stomach"; a closed door is "the door stays SHUT and MOTIONLESS the entire take".
Two reference rules that break generations more than anything else: (1) every @tag appears EXACTLY ONCE, inside ACTIVE REFERENCES — a duplicated tag at the end of a prompt is the most common reason a generation refuses to launch. (2) A location reference carries an explicit line that it controls geometry, materials, light and atmosphere ONLY — never framing; without it the model hands back a near-copy of your plate.
Budget per generation: 9 images, 3 videos and 3 audio references — that budget decides how many named heroes can share a shot.

Optics (the degree ladder) ○ Built-in default

Lenses written in degrees off a ten-anchor ladder; content decides the lens; one lens per shot; a long lens needs its whole observation pattern.

↳ feeds: CINEDANCE (OPTICS)

Geography — master shot & spatial map ○ Built-in default

Open every scene with a ~1s master; paste an unchanged floor-plan spatial map into every shot; tie bodies to visible landmarks; name the 180° line.

↳ feeds: Set Map · CINEDANCE (LOCATION MAP / FIRST FRAME)

GEOGRAPHY — every scene OPENS with a MASTER SHOT: a wide, fixed blocking, about one second, no lines and no action, so the model photographs the arrangement — who stands where, what lies where, where the light comes from — and holds it in every following shot. Remove that second and the heroes start swapping places. Two hacks: let someone say one short word ("hm") so the model treats the wide as a proper shot; and if the scene answers the previous one, feed the TAIL of the previous clip's line into that first second, so the actor answers the right thing in the right tone and the two clips glue at the seam.
Under the master sits the SPATIAL MAP — a floor plan in a few lines, written ONCE per scene and pasted UNCHANGED into every shot. Name the camera side and the 180° line it never crosses; tie every body to a landmark it can SEE (the second chair row, the stage edge, the door, the lamp); use frame-left / frame-right for sides; state EXACTLY how many people are present and that nobody else is; and after every cut, name again who is where and where they look.
Positions come from what is VISIBLE in the plate, never from meters ("to the left of the hero" means nothing — the model doesn't know where the hero is). Give a static dialogue a CORNER of the room, not the whole room — less space, less choice. When a generation contradicts the real location, re-read the REFERENCE, not the prompt.

Acting (restrained, behaviour, subtext) ○ Built-in default

Naturalistic, restrained screen acting — one motivated behaviour per hero, emotion in the delivery, subtext over signalling; NO enumerated eye-life / blink beats (video models exaggerate spelled-out facial actions into overacting).

↳ feeds: Directing Method · THE ACTING SYSTEM · CINEDANCE (CHARACTER ACTING)

ACTING — locked like the look and the voice, and RESTRAINED above all. Perform to a naturalistic, professional screen-acting standard: play the SUBTEXT and react SMALL. Nothing louder than life. NO overacting, no mugging, no theatrical double-takes, no telegraphed or announced emotion, no cartoon faces, no eyebrow "reaction" acting, no over-gesturing, no everyone-reacts-at-once.
ONE behaviour per hero, adapted to each scene but never changed at the core: give a goal and an obstacle and let a SINGLE coherent state play through the shot — carried mainly by the LINE DELIVERY (a short parenthetical on each line: tone, tempo, volume, intent) and AT MOST one small, motivated physical beat (a glance, a stilled hand, a dropped voice). A behaviour impossible in a scene is transferred, not multiplied. Change HOW he fights across a whole SCENE only if the scene truly turns — never several emotional switches inside one short shot.
PHYSICS over adjectives, but SPARING: rather than name an emotion, give ONE concrete muscle cue if it is needed (a tightened jaw, a held breath) — never a list of them. Optionally ONE line of unspoken inner thought, marked never voiced; let the model build the rest from the goal.
NEVER enumerate eye-life: no blink counts or patterns, no "micro-saccades", no pupil dilation, no phased or double-blinks, no "a visible micro-event every N seconds", no eye-rolls or deadpan snaps written as beats. Video models LITERALISE and EXAGGERATE every facial action you spell out — that is exactly what reads as overacting — so name a facial action ONLY when it is the single real beat of the shot, and describe it plainly and small. Steady, lived-in eyes are the default; you do not have to write them.
Let emotion CARRY, not switch: a reaction begins a touch before the other's line ends, and a state EASES across a cut rather than snapping to a new face; the hands can stay quietly busy, the strongest accent being the moment that work STOPS because of what he just heard.

House rules (your own) ○ Built-in default

Any extra production-method rules you want kept. Starts empty — grow it as you learn what to lock in.

↳ feeds: All method-aware apps

Viblu AI Film Studio Dashboard

Film Method KB

📐 Film Method KB