Skip to content

Latest commit

 

History

History
490 lines (367 loc) · 30.9 KB

File metadata and controls

490 lines (367 loc) · 30.9 KB

Higgsfield Vocabulary Reference

Camera Movement Terminology

Linear Movements

  • Dolly In / Out — smooth track toward or away from subject
  • Dolly Left / Right — smooth lateral track
  • Dolly Zoom In / Out — simultaneous dolly + counter-zoom (Vertigo/Hitchcock effect)
  • Super Dolly In / Out — exaggerated fast version of dolly
  • Truck — another term for lateral dolly movement

Vertical Movements

  • Crane Up / Down — vertical rise or descent
  • Crane Over The Head — directly overhead, top-down
  • Levitation — smooth dreamlike float upward
  • Tilt Up / Down — camera rotates on horizontal axis (not physical movement)

Orbit / Arc

  • 360 Orbit — complete circle around subject
  • Arc — semi-circular sweep
  • Lazy Susan — slow turntable rotation
  • Robo Arm — precision mechanical arc path

Zoom

  • Crash Zoom In / Out — rapid sudden zoom
  • Rack Focus — refocus between near and far subjects

Follow / Immersive

  • Action Run — low follow behind running subject
  • FPV Drone — fast agile aerial weaving
  • Handheld — organic shaky realistic movement
  • Head Tracking — locked to character's head
  • Snorricam — mounted on actor, background sways

Cinematic Techniques

  • Bullet Time — slow-motion sweep around frozen subject
  • Dutch Angle — tilted diagonal composition
  • Fisheye — wide distortion lens
  • Whip Pan — fast blur pan
  • Overhead — direct top-down
  • Flying — free-floating aerial glide

Camera Angles

  • Low Angle — camera looks up at subject (power, dominance)
  • High Angle — camera looks down at subject (vulnerability, smallness)
  • Eye Level — neutral, conversational
  • Bird's-Eye View — directly overhead
  • Worm's-Eye View — extreme low, looking straight up
  • Ground Level — camera on the ground surface
  • Canted Angle / Dutch Tilt — tilted horizon for unease/tension
  • Static Oblique — angled perspective, off-axis
  • Over-the-Shoulder (OTS) — conversation framing, shot-reverse-shot
  • POV / First Person — camera IS the character's eyes

Time-Based

  • Hyperlapse — moving camera + time-lapse
  • Timelapse Human / Landscape — fixed camera, fast-forward time
  • Low Shutter — motion blur from slow shutter speed

Cut & Continuity Vocabulary

  • Double contrast — every cut changes BOTH shot size AND camera character (handheld / static / stabilized tracking / crane / aerial). Never repeat the same mode across a cut.
  • Camera character — the mode of motion itself: Handheld | Static/locked-off | Stabilized tracking | Crane/vertical | Aerial/drone
  • Re-anchoring — after any cut returning to established space, re-state who is where and which direction they face. If a character moved left-to-right before the cut, same direction after.
  • 180° rule — keep the camera on one side of the imaginary line between subjects to preserve screen direction; crossing it flips left/right and disorients the viewer
  • Exit-frame = implicit cut — once a character leaves frame, they are gone for the remainder of that shot. Never choreograph exit + re-entry in the same continuous shot.
  • Causally-motivated insert — sub-second (0.3–0.5s) static detail the viewer understands WHY they see. Always name the subject ("HIS hand gripping metal"), never generic ("a hand").
  • Match cut — cut between two shots linked by shape, motion, or composition
  • Anchor-gesture cut ("cut on action") — to bridge two separately-generated clips so they cut together cleanly, end the outgoing clip and begin the incoming clip on the same repeated gesture (e.g. an ear-cup tap, a door push, a head turn). The reused motion lets independently-rendered scenes "cut on action" most of the time — the editor's match-on-action, engineered at prompt time. Write the identical gesture into the last CUT of scene N and the first CUT of scene N+1.
  • Multi-take micro-slice assembly — a finished scene is often cut from many takes, not one keeper clip: the spoon from one generation, the flame from another, the pour from a third, the sip from a fourth. Plan for it — scrub each take for its best 1–3 seconds, bank the slices, and assemble. This is the take-selection face of the acceptance-rate discipline (DISCIPLINE.md); see also skills/higgsfield-seedance/FAILURE-MODES.md § Failed-generation salvage.
  • Smash cut — abrupt, jarring cut with no transitional softening
  • Hard cut / direct cut — standard straight cut, no effect
  • L-cut / J-cut — audio from next shot leads picture (L) or vice versa (J). The L-cut audio bridge is the production-team's working name: hold the audio of the outgoing shot across the cut so the incoming visual lands inside an already-established soundscape.
  • Whip-pan transition — blur pan that bridges two shots as a single motion
  • Camera-relative-to-previous-shot continuity — when planning the next shot, name the camera position relative to the previous shot's camera (camera now behind him, facing the same direction he was facing) rather than absolutely. Keeps the cut from disorienting the viewer.

Editing Syntax

Notation conventions for time-based and editorial structure inside prompts.

  • [0-Ns] bracket notation — time-range prefix for shot phases inside a single continuous shot. Example: [0-3s] subject enters frame, [3-7s] camera pushes in. Used for per-second beats inside single-shot continuous motion (see skills/higgsfield-seedance/SKILL.md § Single-vs-multi-shot decision).
  • Establishing → phrase → reaction arrow notation — editorial-flow notation for multi-cut dialogue or beat sequences. Reads as "establishing shot, then phrase/dialogue, then reaction shot." Compact prompt-side shorthand for a three-beat cut structure.

Motion Hierarchy

The 4-layer motion hierarchy separates simultaneous motions a shot can carry so the prompt names each layer explicitly:

  1. Subject motion — what the character/object does (walks, turns, lifts hand)
  2. Internal motion — micro-motion within the subject (breath, hair drift, fabric, eye motion, micro-expression)
  3. Camera motion — what the camera does (pan, dolly, push-in, hold)
  4. Environmental motion — what the world does (rain, leaves, traffic in background, light shifting)

Specifying each layer separately prevents the multi-motion overload failure mode (see skills/higgsfield-seedance/FAILURE-MODES.md § Multi-motion camera overload).

Camera Contract

A camera contract is the prompt-side discipline of stating camera behavior as an explicit rule before describing subject or action. When camera behavior is left implicit, the model defaults to its training-prior interpretation, which may not match the intended shot — stating the contract anchors the model's output.

Three example contracts, drawn from common shot shapes:

  • Static locked-off camera. Zero movement. No pan, no zoom, no dolly, no shake. — atmospheric / observational shots, product reveals, sacred-register imagery
  • Slow push-in only — 10% scale change over the full duration. — quiet emotional intensification, realization beats, single-subject scenes
  • Single handheld drift, slight organic sway, no cuts. — naturalistic / documentary register, intimate scenes, low-fi visual identity

Pair every camera contract with negative-prompt reinforcement (see § Composition Vocabulary → Negative-prompt reinforcement below) — name the excluded camera moves in the negative prompt so the model is constrained from both sides.

Part of the 5-pillar cinematic-motion-language framework, see also § Motion Physics Anchor (below), § Lens Behavior Sequence (below), § Composition Vocabulary → Spatial Zoning and Negative space. Translated from Adil Aliyev's cinematic-motion-language.md source corpus.

Motion Physics Anchor

A motion physics anchor specifies how fast a moving element moves through the frame by anchoring the speed to a physical analogy or a time measurement.

Distinct from § Subject & Character → Movement Quality (further below), which describes the character of motion — fluid, jerky, hesitant, confident — performance direction rather than speed specification. The two vocabularies compose: a character can move with stumbling quality (Movement Quality) at the pace of a clock's hour hand (Motion Physics Anchor). Different axes of motion description.

Physical-analogy speed references anchor abstract pace to observable real-world motion:

  • like dust suspended in honey — extremely slow, viscous, deliberate
  • like embers floating in still air — slow, weightless, drifting
  • like smoke through a cathedral at dawn — slow, layered, atmospheric
  • like the surface of a lake disturbed by a single drop — slow ripple-outward expansion
  • like a flag in a steady cross-breeze — moderate, periodic, directional
  • like a slammed door rebounding — fast, decaying oscillation

Time-anchored measurements specify pace as a quantity:

  • one full revolution across the entire 10-second clip
  • roughly 6 degrees per second
  • the pace of a clock's hour hand — imperceptibly slow
  • travels the full arc in 8 seconds with no pause

When motion-speed precision matters, anchor to physics analogies or time measurements rather than adjectives. Adjectives like slow or fast work as quick-spec for casual prompts but are less reliably interpreted than physical analogies or time anchors. The directional pattern: more specific anchor → more controlled output.

Part of the 5-pillar cinematic-motion-language framework, see also § Camera Contract (above), § Lens Behavior Sequence (below), § Composition Vocabulary → Spatial Zoning and Negative space. Same translation discipline as vocab.md § Scene-physics lighting (replace adjectival descriptors with named physical mechanisms). Translated from Adil Aliyev's cinematic-motion-language.md source corpus.

Lens Behavior Sequence

A lens behavior sequence describes a focus or depth-of-field event as a narrative with structure: trigger → shift → state → return → repeat. Models tend to produce cleaner / more reliable focus events when the sequence is described as cause and effect rather than as a single static depth-of-field state.

Example sequence:

Focus opens on the subject. As the foreground element crosses the lens plane, focus shifts onto it — the subject softens into warm bokeh. The element drifts past. Focus breathes back to the subject. This cycle repeats organically 2-3 times.

Key vocabulary:

  • shallow depth of field — narrow plane of sharpness, foreground/background fall off
  • focus-breathing — organic in/out focus shift between two subjects without a hard rack
  • rack focus — deliberate, directional focus shift between two named subjects (see § Camera Movement Terminology → Zoom → Rack Focus above for the bare-bones entry)
  • bokeh silhouette — out-of-focus subject reduced to a soft warm shape against background light
  • lens plane crossing — the moment a foreground element passes between camera and subject, triggering a focus event
  • anamorphic lens rendering — oval bokeh, horizontal flare character, widescreen feel (see also § Visual Style Vocabulary → Named Platform Styles → Anamorphic for the style register, and § Aspect Ratio: output spec vs. style register for the output-vs-style boundary)

Part of the 5-pillar cinematic-motion-language framework, see also § Camera Contract (above), § Motion Physics Anchor (above), § Composition Vocabulary → Spatial Zoning and Negative space. Translated from Adil Aliyev's cinematic-motion-language.md source corpus.


Shot Size Vocabulary

Term Frame content
Extreme Long Shot (ELS) Vast landscape, subject tiny or absent
Extreme Wide (EWS) Subject tiny in environment
Wide Shot / Long Shot (WS/LS) Full body + surroundings
Medium Long Shot / Cowboy (MLS) Subject from mid-thigh up
Medium Wide (MWS) Subject from knees up
Medium (MS) Waist up
Medium Close-Up (MCU) Chest up
Close-Up (CU) Face
Extreme Close-Up (ECU) Detail — eyes, hands, object
Insert Shot Detail of an object or action
Over-the-Shoulder (OTS) Looking over a character's shoulder
POV First-person character perspective
Two-Shot Two characters in frame
Cowboy Shot Mid-thigh up, posture and readiness

Visual Style Vocabulary

Named Platform Styles

  • Cinematic — polished, high-contrast, professional
  • VHS — retro grain, color bleed, analog warmth
  • Super 8MM — warm film grain, soft vignette, nostalgic
  • Anamorphic — ultra-wide, lens flares, epic scale
  • Abstract — surreal, non-representational, artistic

Aspect Ratio: output spec vs. style register

Concern What it is Where it belongs in a prompt Bound by
Output ratio Hard platform spec — what the model actually emits Header (e.g. "Aspect ratio: 16:9") Per-model enum (Kling 3.0: 16:9 / 9:16 / 1:1; check higgsfield model get <model>)
Anamorphic / 2.35:1 / 2.39:1 / Scope / Cinemascope Cinematography register — anamorphic flares, letterboxed framing aesthetic Look line, as a style request ("anamorphic-style flares, letterboxed composition") Stylistic — the model renders the look within the chosen output ratio

Conflating these produces incoherent prompts. "16:9 anamorphic" requests two contradictory things at once. The model will either ignore one or produce uneven results. Separate them.

Color Grade Language

Mood Description
Blockbuster Teal shadows, orange highlights, high contrast
Cold thriller Desaturated blue-grey, crushed blacks
Warm nostalgia Golden amber, lifted shadows, soft grain
Cyberpunk Neon magenta + cyan, deep shadows, HDR
Horror Sickly yellow-green, low contrast, murky
Romance Soft pink-gold, lifted shadows, dreamy
Documentary Natural neutral, no grade
Epic fantasy Deep jewel tones, volumetric light
Noir High contrast black and white
Post-apocalyptic Desaturated orange-brown, dust haze

Color rules

  • 60/30/10 color rule — production-direction color-palette discipline. A frame's color reads cleanest when 60% is one dominant color, 30% is a secondary, and 10% is an accent. Prompts that name a single dominant tone + supporting palette land more cohesively than prompts that list five competing colors.

Film Stock Emulation

  • Kodak Portra 400 — warm, rich skin tones, slight grain
  • Fuji Velvia — vivid saturated colors, fine grain
  • Kodak Vision3 500T — cinematic, natural, slight warmth
  • Ilford HP5 — classic black and white, visible grain
  • Kodak Ektachrome — bright, contrasty, clean slide film

Lighting Vocabulary

Type Description
Golden hour / Magic hour Warm directional, sun near horizon
Blue hour Cool, soft, just after sunset
Overcast / Softbox Diffused, soft, no shadows
Neon Colored artificial from signs/screens
Volumetric Light rays visible (fog/dust/smoke)
Practical only Light from sources visible in frame
Side-lit Single strong source from one side
Backlit / Rimlit Light source behind subject
Motivated Light that matches a logical source in scene
Hard light Strong single source, defined shadows
Soft light Diffused, wrapped, minimal shadows
Rembrandt Triangle of light on shadowed cheek
Butterfly / Paramount Overhead, shadow under nose (glamour)
Split lighting Half face lit, half in shadow
Chiaroscuro Extreme light/dark contrast
High-key Bright, minimal shadows (comedy, commercial)
Low-key Deep shadows, moody (noir, horror)
Harsh midday sun Hard overhead, strong defined shadows

Scene-physics lighting

Lighting described as a physical event in the scene rather than as a stylistic adjective. State the source position, the surfaces it reflects off, and the resulting shadow geometry — the model uses each named piece to reconstruct illumination consistently across shots.

  • Source positionkey light from screen-left, ~30° above eye line
  • Reflection / bouncewarm bounce off the wooden floor lifting under-jaw shadow
  • Shadow geometryhard shadow on screen-right wall reading the silhouette of the desk lamp
  • Ambient densitydense ambient haze in the foreground softening contrast 6m+ out from camera

The scene-physics framing replaces stylistic adjectives (moody, dramatic, cinematic) with named physical mechanisms that compose. The model can render the named mechanisms; it can only approximate the adjectives.

When integrating a preserved real subject or an added creature into a new plate (video-to-video), color matching alone reads as "pasted in." Match key direction, environmental bounce, optics/atmospheric haze, and edge grounding together — the full recipe is in skills/higgsfield-seedance-vfx/SKILL.md § Lighting integration.


Subject & Character Vocabulary

Body Language / Posture

  • Upright, tense, coiled, relaxed, hunched, squared-off
  • Arms open / crossed / raised / at sides
  • Weight forward / back / centered
  • Eyes wide / narrowed / cast down / direct to camera

Movement Quality

  • Fluid, jerky, precise, hesitant, confident, stumbling
  • Slow-motion, overcranked (high fps = slow playback)
  • Undercranked (low fps = fast playback)
  • Natural, stilted, mechanical, animal-like

Emotion Cues (for camera and performance)

  • Realization: slow Dolly In, eyes widening
  • Fear: Dutch Angle, Handheld
  • Power: Crane Up, 360 Orbit, wide stance
  • Vulnerability: Overhead, Dolly Out, small in frame
  • Intimacy: Dolly In, shallow depth of field
  • Chaos: FPV Drone, Handheld, fast cuts

Emotion as Visible Behavior — Channels

Eight named channels through which emotion becomes visible to the camera. Use these to decompose abstract emotion language ("she is afraid," "he is angry") into observable physical behavior the model can stage.

  • Breath — rate, depth, hold, exhale character
  • Eye behavior — fixation, blink rate, gaze direction
  • Jaw tension — clenched, slack, working, set
  • Loss of focus — gaze drifting, attention slipping, response delay
  • Scanning pattern — eye/head sweep direction, tracked target
  • Delayed recovery — lag between stimulus and adjustment back to baseline
  • Control attempt — visible suppression or regulation effort
  • Emotional residue after the peak — what reads on the body after the dominant beat

The Micro-Expression Vocabulary section below catalogs named endpoint expressions (Suppressed Smile, Cold Calculation, etc.) — those are combinations of these channels. The channels above are the substrate; the named expressions are what the substrate produces.

These eight channels are the behavioral decomposition of emotion. The anatomical decomposition — named facial muscles as FACS Action Unit codes (AU6 cheek raiser, AU12 lip-corner puller) for muscle-level Seedance control — is its sibling, in skills/higgsfield-facs/SKILL.md. Channels say what the body does; AUs say which muscles move.


Environment & Atmosphere Vocabulary

Time of Day

  • Dawn — first light, mist, cool blue to warm orange
  • Golden hour — 1hr after sunrise / before sunset
  • Midday — harsh overhead light, deep shadows
  • Overcast day — flat, diffused, neutral
  • Blue hour — after sunset, cool soft glow
  • Night — dark, pools of light, shadow-heavy

Weather / Atmosphere

  • Rain: wet surfaces, reflections, steam rising
  • Fog/mist: soft depth, reduced contrast, mystery
  • Snow: muted colors, diffused light, silence implied
  • Heat haze: shimmering air, warm distortion
  • Wind: moving elements, hair/fabric/trees
  • Dust/smoke: volumetric light opportunities

Environment Types

  • Urban: concrete, glass, neon, crowds, traffic
  • Industrial: metal, rust, machinery, steam pipes
  • Natural: organic textures, varied light, scale
  • Interior: controlled light, props, intimacy
  • Wilderness: raw, unpredictable, vast
  • Fantasy: designed, non-real, heightened

World Through Recurrence

Eight named substrate axes through which a world becomes legible to the audience across multiple shots. Use these to identify what should repeat consistently across scenes — a world emerges from what recurs, not from what gets explained.

  • Recurring spaces — locations that return, space-types echoed across acts
  • Light behavior — how light enters, falls, decays in repeated patterns
  • Silence — the texture and timing of quiet, what gets withheld between beats
  • Materials — surface vocabulary, substance choices, tactile signature
  • Architecture — corridor logic, threshold use, the way built form constrains motion
  • Props — physical objects that reappear with consistent meaning
  • Emotional conditions — recurring affective states, dread density, what lingers after a beat resolves
  • Camera distance / rhythm — proximity choices, when to step back, pacing patterns

Audio / Sound Vocabulary (for Kling 3.0 and audio-capable models)

Sound type Description for prompt
Ambient "soft city hum", "distant traffic", "forest insects"
Weather "rain on cobblestones", "wind through trees", "thunder rumble"
Movement "footsteps on gravel", "fabric movement", "glass crunch"
Mechanical "engine revving", "metal scrape", "hydraulic hiss"
Human "breathing", "heartbeat rising", "crowd murmur"
Music cue "tense strings build", "low bass drone", "electronic pulse"

Platform Feature Vocabulary

Term Definition
Soul ID Upload 20+ photos of a real person to train identity consistency across generations
Soul Cast Generate AI actors from parameters (no photos needed) — Cinema Studio 2.5 feature, powered by Nano Banana 2
Soul HEX Extract color palettes from reference photos for brand-consistent, color-matched visuals
Soul Cinema Preview Higgsfield proprietary cinematic-grade image model, prompt-driven only, excels at close-ups
Elements 3.0 Reference system using @element_name syntax for cross-shot subject consistency (Kling 3.0)
Voice Binding Lock specific voice profiles to specific characters across shots (Kling 3.0)
Performance Cloning Act out a scene on camera → AI re-renders preserving likeness and voice (Kling 3.0 Omni)
Motion Control Transfer motion from reference video to new character/scene (Kling 3.0 Motion Control)
Reference Anchor Cinema Studio system that locks character geometry across all shots
Hero Frame Key image generated before video to define visual tone, lighting, and composition

Image Reference Notations — @Image1 vs <<<image_1>>>

Both notations refer to the same underlying @-reference mechanic; the bracket style is contextual to which surface is being addressed.

  • @Image1 — the platform-side reference syntax used in actual prompts to the Higgsfield generation engine. Capital I, no underscore, single @. Used across templates/ and most sub-skill prompt examples (e.g., templates/02-product-ugc-showcase.md, MODELS-DEEP-REFERENCE.md § The Rule of 12). Pattern in prompts: @Image1 as the character, @Image2 as the background.
  • <<<image_1>>> — the Seedance 2 Skill's internal explicit-reference notation, used in user requests to the docs/Seedance 2 Skill.md bilingual EN+ZH prompt director. Lowercase image, underscore, triple angle bracket. Recognized by the Seedance Skill when typed in a user request, then mapped to the appropriate scene element (per docs/Seedance 2 Skill.md:178 § Image reference system).

Per-slot role convention — production practice locks a stable role assignment per @Image slot, kept identical across every prompt in a shot list: @Image1 = character identity, @Image2 = costume, @Image3 = environment + lighting, @Image4 = composition, @Video1 = motion, @Video2 = camera movement, @Audio1 = rhythm + atmosphere. The convention is team-side discipline, not model-enforced; the payoff is reference-stability across long shot lists (see skills/higgsfield-seedance/SKILL.md § Per-Image Role Convention).


Composition Vocabulary

Vocabulary for what's IN the frame and where it sits — distinct from camera-side language (which describes how the camera moves) and lighting (which describes illumination).

Negative-prompt reinforcement

A cross-cutting compositional discipline: when the prompt declares a positive constraint on the frame — a locked camera, a black zone, a deliberately-empty region — mirror the exclusion in the negative prompt so the model is constrained from both sides.

Example pairs:

  • Positive: Static locked-off camera. Zero movement. Negative: camera movement, pan, zoom, dolly, shake
  • Positive: Left third: pure black, no motion, no light. Negative: particles on the left side, light on the left side, movement on the left side
  • Positive: Foreground plane: particle layer only — no subject. Negative: subject in foreground, character in foreground

The positive-side declaration names the intent; the negative-side reinforcement prevents the model from filling against the named intent. Pillars that use this technique: § Camera Movement Terminology → Camera Contract (camera-rule reinforcement); § Spatial Zoning below (zone-rule reinforcement); § Negative space below (excluded-region reinforcement).

Spatial Zoning

A spatial zoning prompt divides the frame into named regions and assigns explicit rules to each region. This prevents the model from filling empty space with invented content — a falsifiable production claim, well-aligned with how transformer-based video models respond to explicit constraints.

Region naming conventions (combine as needed):

  • Thirds: left third, center third, right third
  • Depth planes: foreground plane, midground, background
  • Halves: upper half, lower half
  • Asymmetric splits: right two-thirds / left void, left two-thirds / right void

Zone-rule examples (each region declares its content and behavior):

  • Left third: pure black, no light, no particles, no movement.
  • Right two-thirds: all action contained here.
  • Foreground plane: particle layer only — no subject.
  • Background: unlit void, no detail.

Always reinforce zone rules in the negative prompt — see § Negative-prompt reinforcement above. § Negative space (below) is one common zone type within a spatial-zoning system.

Part of the 5-pillar cinematic-motion-language framework, see also § Camera Movement Terminology → Camera Contract, § Motion Physics Anchor, § Lens Behavior Sequence, § Negative space (below). Translated from Adil Aliyev's cinematic-motion-language.md source corpus.

Negative space

The deliberately-empty area of the frame that isolates or weighs the subject. Production-direction language: negative space sits center-left of frame, narrowing as the character advances. Naming negative space as a compositional element prevents the model from filling it with incidental detail.

When the prompt names a negative-space zone, mirror the exclusion in the negative prompt — see § Negative-prompt reinforcement above. Negative space is one named zone type within a § Spatial Zoning system: Spatial Zoning declares the region rules; negative space names which regions are deliberately empty. The two compose.

Source register sometimes names negative space evocatively: sacred emptiness, active darkness, not background, deliberate compositional weight. Production-direction register tolerates evocative naming when the evocation is grounded in a compositional rule — e.g., naming a zone sacred emptiness after the zone has been declared as no-light, no-motion, no-particles. Without the underlying rule, evocative naming reads as adjective-stacking; the rule is what makes the evocation operative. (Evocative naming examples sourced from Adil Aliyev's cinematic-motion-language.md.)

Part of the 5-pillar cinematic-motion-language framework, see also § Camera Movement Terminology → Camera Contract, § Motion Physics Anchor, § Lens Behavior Sequence, § Spatial Zoning (above).

Crossing rule

Film-grammar clause for whether characters may cross positions on-screen within a shot or across a cut. Distinct from the camera-side 180° rule (which is about the imaginary line between subjects): the crossing rule names what subjects may do, the 180° rule names what the camera may do. Example: Character A stays on the left, Character B stays on the right, neither crosses the central vertical axis.

Coordinate notation

Paired qualitative + percentage spatial anchors for in-frame placement. Character A stands in the right third, x-position 70%, y-position 50%, frame occupancy 25%. The qualitative term gives the film-language hook; the percentage gives the precision target. See skills/higgsfield-seedance/SKILL.md § Frame Coordinate System for the full coordinate system + caveats.


Production Vocabulary

Terminology imported from traditional film production into AI-cinema practice — names for disciplines that recur in production workflow.

  • Script supervising — the traditional-film-production discipline of tracking every character's state (injuries, clothing, props, hair) across shots so continuity holds across cuts. Imported into AI-cinema via per-state character anchor sheets: a character with five stages (e.g., initial / fight-injuries / partially-transformed / almost-fully / completely-transformed) gets five distinct Soul ID sheets, one per stage. See skills/higgsfield-soul/SKILL.md § Multi-Form State Tracking.
  • State lock — the per-shot directive that names the character's current emotional or physical state (calm, exhausted, injured, soaked, in motion) inside the Character Anchor Block. Distinct from pose (physical configuration) and facial expression (emotional register): state lock names the persistent condition that should hold across the shot's duration. See skills/higgsfield-soul/SKILL.md § Character Anchor Block.

Micro-Expression Vocabulary

The Emotion as Visible Behavior — Channels section above catalogs the eight substrate channels these endpoint expressions decompose into.

Performance directions for Soul Cast actors and character prompting:

Expression Description
Deadpan Neutral Flat affect, no visible emotion
Fierce Focus Intense locked gaze, total attention
Suppressed Smile Fighting back a grin, corner of mouth twitching
Quiet Devastation Eyes glassy, jaw tight, holding it together
Cold Calculation Eyes scanning, no emotional leakage, clinical
Bitter Amusement One-sided smirk, eyes not smiling
Frozen Shock Mouth slightly open, eyes fixed, body still
Simmering Rage Clenched jaw, flared nostrils, steady stare
Vulnerable Openness Soft eyes, slightly parted lips, unguarded