Higgsfield Acting

Higgsfield Acting Director

[OFFICIAL — Higgsfield "Hell Grind" open-source brief] — the performance system behind Higgsfield's 95-minute AI feature film, adapted to this repo's prompt surfaces. That tag covers the untagged text only: § The layer above the pillars, the silent-listener task, the eye-task paragraph and the voice-bible note each carry their own tag and are not from the brief (label legend: ../shared/provenance.md).

The core axiom of the entire system: acting is BEHAVIOR under pressure, not a display of emotion. A character wants something, something is in the way, and they act to get it. Emotion is a byproduct of that struggle — never the thing you write directly.

This is a layer on top of ../higgsfield-seedance/SKILL.md: the paragraph this skill produces goes into the PERFORMANCE / CHARACTER ACTING block of an otherwise normal Seedance prompt. It does not change camera, light, wardrobe, or grade.

QUICK FACTS

Generated-checked block (scripts/build_index.py verifies anchors). Routing aids — read the linked sections for the rules.

• Write the objective (a verb aimed at a partner), never the state — "make him confess", never "be angry" →
• 2–4 beat changes per scene, each one visible in behavior: a pause, a posture change, a tempo change, a shift of gaze. Unchanged behavior for the whole shot = played flat →
• Listening beats speaking. The reaction starts before the partner's line ends; a hard question earns a micro-pause before the answer →
• The body carries the biography: center of gravity, tempo, openness, breath — set all four before any psychology →
• Business — a physical task the hands are doing — kills theatricality; the strongest accent in a scene is the moment the character stops it →
• Distance is drama; a change of distance is a beat change. Status is what you DO, and status breaks are the most interesting thing in a performance →
• The master profile is 150–220 words, one paragraph, fixed block order, written once per character and then rewritten per scene — never pasted →
• Every tic carries a trigger; every mask carries a crack — at least one "However, when X…" clause per profile →
• Eye life is mandatory and never optional — give the eyes a task aimed at the partner, then saccades, blink quality, eyes-lead-thought; catchlights only make the task legible, they never fix a dead stare. Dead eyes are the number-one tell of AI acting →
• Scene adaptation transforms, never deletes: a behavior that can't physically happen is displaced into another outlet, not removed →
• The voice prompt is locked — one per character, from one voice bible, never adapted per scene; on 2.0 pasted verbatim into the audio field each time they speak, on 2.5 written in the character's role sentence, not the audio field, because the reused sheet carries the voice (once per prompt or once per project is OPEN, no default — P2-2) →
• States, not transitions. Models fail process and nail state: "mid-throw, arm extended", not "reaches in, pulls out, winds up" — for a peak reached through a reversing process; simple same-direction motion that must fill the clip is chained instead →
• Ensemble reactions travel in a wave, never in sync; the strong are still and quiet, the weak fidget and shout →
• 15 named bad-acting symptoms with prompt-level fixes, and a 0–5 self-check scale — aim every hero shot at 4+ →

Where this sits among the repo's performance tools

Use this skill to decide what the body does; use FACS to specify which muscles do it in a close-up. They compose — a FACS beat schedule inside a paragraph written by these rules is the highest-resolution performance the platform supports.

Part I — The craft

What good acting is

Truthful behavior under imaginary circumstances. Not depicted emotions, not recited text, not "expressive faces."

The five pillars of every scene

Every character in every scene decomposes into five elements. If one is missing, the performance falls apart.

1. Objective. What the character wants IN THIS SCENE, RIGHT NOW, FROM A SPECIFIC PERSON. Always a verb aimed at the partner: make him confess · beg a week's extension · convince her I'm not afraid. Never a state ("be angry", "feel guilty") — states cannot be played. Behind the scene objective sits a super-objective: what the character wants across the whole story, with every scene objective a step toward it.

2. Obstacle and stakes. What prevents them — external (another character wants the opposite; witnesses in the room; two hours to deadline) or internal (pride won't let them beg; they don't believe their own words). Always answer what happens if I do NOT get what I want? — and the answer must frighten the character. The higher the cost of failure, the more taut the scene.

3. Tactics. The concrete method of pursuing the objective right now, as an action verb: press · charm · shame · plead · provoke · bargain · threaten · stall. When a tactic fails, a living person changes it. One tactic for a whole scene is dead acting.

4. Beats. The smallest unit of action: the stretch during which the character wants one thing and pursues it one way. A beat ends when the objective is achieved, the tactic fails, new information arrives, or the balance of power shifts. Every beat change must be VISIBLE in behavior — a pause, a change of posture, a change of speech tempo, a shift of gaze. A good scene has 2–4 beat changes.

5. Subtext. What the character actually thinks and wants, as opposed to what they say. Subtext is not performed — it leaks out on its own when the character plays the true objective while speaking the false text. Buildable markers: questions that aren't questions · repetitions (asking the same thing — they don't believe the answer) · abrupt topic changes · jokes at the wrong moment (a shield against vulnerability) · answers that are too short ("Fine." "Sure." — a closed door).

The layer above the pillars — one direction, different fuel

[EMPIRICAL — Tigran's tig-acting-task skill (2026-07-10), re-derived 2026-08-22] The five pillars are per-character. They do not say what holds an ensemble together, and that is why scenes built from them alone can read as several good performances that are not in the same scene. One layer sits above them.

The scene has ONE direction, and every character plays toward it. Usually unspoken — a mutual silent agreement about how this time will be lived. Part without pain, stay positive (a mother packing her son's kit before he leaves). It belongs to everyone in the room at once. It is not the film's dramaturgic function: characters never play the reveal or the theme, which are accomplished through them as a byproduct.

Each character pushes that direction for their own reason — the MOTIVE. Same vector, different fuel. The son keeps it painless for his mother; the mother out of superstition (tears before a journey are a bad omen). The fuel is what makes each performance distinct while the scene still reads unified — and it is the piece most often skipped, which is what collapses an ensemble into one note repeated.

Given circumstances constrain the motive: a character who already took a compromised job cannot play moral innocence. If the motive contradicts the backstory, re-derive it.

Distinguish carefully, because these are four different things and only the last two are per-character:

Name the event from the ENDING

Read how the scene ends before naming what it is about. The last line or beat is the key you read the whole scene backward through — and watch for the double-meaning last line, spoken about one thing and meant about another ("Poor bastard. Just can't forgive himself" — said over a patient, meant about the speaker).

The test: the event must contain EVERY character in the scene, including silent and unconscious ones, as participants or mirrors of the same process. If a character stands outside the named event, the event is named wrong — rename it until they are all inside it. This is the cheapest structural check in this file and it catches the scene where one character is merely present.

The physical action is the channel

The surface activity — the terrain — stays as the physical action, and each character pursues the event through it, via their own distinct, camera-readable behaviour. The invisible task must have a visible channel or the model has nothing to render.

One terrain (routine hospital rounds), one event (the search for self-forgiveness), three channels:

• one character through remembering — REM under closed lids, a tear
• one through obligated actions done right — correct metrics, on time, double-checking an entry
• one through caring — patient-care gestures beyond the checklist, checking the man rather than the chart

Give every character in the scene their own physical channel for the same event. Different behaviours, one event, one terrain. This is what turns an acting note into something a video model can actually generate.

Contrast pairing — build duos on mirrored +/−

A two-hander is richer when the pair is built on contrast: each character carries one plus and one minus, inverted relative to the partner, and one axis is named as the essential axis — the opposition the audience actually reads. The other traits are colour.

A lab pair, essential axis care for the patient:

• Doctor — loyal (+) / careless (−). Genuinely loyal, but to the institution and the money. His plus serves the wrong master, so on the essential axis he is minus.
• Medic — compromised (−) / caring (+). Compromised by taking the job at all, but caring toward the patient. His minus is real; his plus survives inside it — on the essential axis he is plus.

Both still push the same scene direction. The contrast lives under it and leaks out through the tactics: the careless one's routine is real, the caring one's routine is armour, and it cracks.

The seeming trait is often scar tissue over its opposite — "careless" is hope lost, not care absent. Direct the history, not the surface.

Listening and reaction

Performance lives not in the lines but between them. Four observable markers to write in:

1. The reaction starts before the partner's line ends. A person grasps the point mid-phrase — the face and body already answer. A neutral face until the line ends, then "switching on", is dead acting.
2. Thought before word. Before a hard answer there is a micro-pause — the person visibly decides what to say. Instant answers at uniform speed read as memorized text.
3. The assessment moment. After news, a threat, or an insult, the character needs time to digest — from a fraction of a second to a long pause. Cinema lives on close-ups of these assessments: a face, a defocused gaze, the world swimming.
4. Contagion from the partner. Tempo, volume, and energy shift in response: a shout is answered either with a counter-shout or with pointed quiet — but answered, not continued over.

A silent listener still gets a task, not just markers. [EMPIRICAL — v3.33.0 source drop, re-derived 2026-08-22; source file not recorded] The four markers above describe how a reaction reads; they do not give the listener anything to be doing between them, and a listener with no task is where the dead face comes back in a two-shot. Name the work: decide whether he is serious · wait for the opening · protect the mood · catch him in the lie. Every character in frame gets living eyes this way.

The body

Physical state before psychology. Set four parameters for every character before any inner life:

• Center of gravity — high (chest, chin: confidence, aggression, status) or low (shoulders, slouch: fatigue, fear, submission).
• Tempo — fast/ragged (nervousness, stimulant energy) or slow/economical (control, threat — the most dangerous people move least).
• Openness — squared shoulders and open palms, versus crossed arms, dropped head, closed poses.
• Breath — high and rapid (panic) versus low and slow (control). Breath is the most honest indicator of state, and sound must match physics: someone who just ran cannot speak on a steady voice.

Business — the physical task. A character almost always needs a DOING. They don't "have a conversation" — they fix an engine, count money, cook, wipe a glass, and talk over the top of it. Business kills fake theatricality (the hands are busy with truth), creates rhythm (pauses fill with action), and generates subtext (how a person counts money says more than the words).

The interrupted-action rule. The strongest accent is when a character stops the business. If he was slicing bread and stopped at a phrase, the phrase became an event. Use the stop as punctuation.

Proxemics — distance as drama. The distance between characters is a visible graph of their relationship:

A threat whispered 10 cm from a face is scarier than one shouted across a room. Scene drama is often just the story of distance: who closes it, who breaks it, who freezes. A change of distance is a change of beat — and it must be written in metres, per the measurable-language rule in ../higgsfield-seedance/SKILL.md.

Status — the invisible hierarchy. Status is what you DO, not who you are.

• High status: an immobile head, slow movements, long gazes, pauses before answering, taking up space, touching other people's things.
• Low status: fussing, frequent self-touching (face, hair), broken speech, filler-laughter, asking permission with the eyes.

The most interesting thing in a performance is a status break: the boss who shows fear for one second; the underling who suddenly stops smiling. A powerful scene shape is enter high and collapse, or enter low and flip the room.

How good acting sounds

• Rhythm is written in the text — deliver it precisely; fast ≠ mushy.
• Overlaps are normal (lines stepping on each other's tails), but key words stay clean.
• Volume contrast: the most frightening things are said the quietest. Shouting is the currency of the weak; the character who owns the scene lowers the volume and everyone leans in.
• Pauses are events, not holes — a pause is legal only if something happens inside it: an assessment, a decision, a refusal to answer.
• Real speech has litter: interruptions, half-heard words, repetitions, unfinished phrases. Perfectly constructed sentences kill street truth.

Part II — Writing the acting

The acting master profile

Every recurring character gets one master profile — the permanent source of truth about how they act. Written once, then adapted per scene. Target: 150–220 words, one flowing paragraph, fully observable and filmable.

Block order is fixed:

Character acting as [NAME]. [Build, physique, posture — the body as a document of their
biography]. [The psychological engine in one clause — the inner drive that explains the
physicality]. Vocal profile: [pitch/timbre, accent/origin, pace and delivery manner, and how
the voice breaks or shifts under emotion]. Key physical habits and tics: [signature tic with
its trigger; stress tic with its trigger; concealment behavior — what they do to hide what
they feel; the facial mask and the exact condition under which it cracks]. Eye life: [the
eyes' standing task toward others, blink quality and rate, scanning pattern, gaze-before-head].
Walking style: [the gait
as characterization — named and specific, with weight, rhythm, and foot placement]. However,
when [emotional trigger], [the transformation — how the posture, gait, and face change].
[Optional: the softening target — the one person or thing that makes the face genuinely soften].

Rules for each block:

1. Only observable behavior. Every inner state needs a body marker. Never "he is nervous" — write the trembling lower lip, the heavy swallow, the long inhale through the mouth and sharp exhale through pursed lips.
2. Every tic has a trigger. Not "he cracks his knuckles" but "he cracks his knuckles during mundane conversations to fake confidence." Format: [tic] + [when/why]. Tics without triggers are decoration; tics with triggers are dramaturgy.
3. Name the gait. A coined name in quotes anchors the biomechanics — a "power-walk", a "peacock strut", a "gallery walk", a "battering-ram stride", a "dreadnought pace" — then unpack it: weight, step, what the torso and arms do, what the head does.
4. Build in the mask AND the crack. The single most cinematic device in a profile is the conditional transformation: the facade, plus the precise trigger that collapses it. A dominant swagger that crumbles into a childlike slouch when rejected; stern discipline that melts into a radiant smile at one specific person. Every profile carries at least one "However, when X — …" clause. A character playing two truths at once is the difference between a puppet and a person.
5. One softening target. Where it fits, give the face exactly one person, animal, or object it genuinely softens for. One — not two. This humanizes without diluting.
6. No wardrobe. Clothing lives in the scene/look block, never in the acting profile. The profile must survive any costume change.
7. No camera, no color. Acting drives performance, face, voice, and motion only.
8. Physique carries biography. Build and posture tell the backstory — a build "weathered from past struggles", shoulders "perpetually tense", a posture "over-corrected to project authority". Profession, past injuries, and self-image must be readable in the body.

Age is not a profile field. Engine rule 1 (../higgsfield-seedance/ENGINE-RULES.md) is age-blind characters, and the content filter tightens sharply on age words. Carry the same information through build, wear, posture, and voice — "a body that has already been repaired twice", "a working-class rasp", "an old boxer's walk" — never a number and never boy / girl / young / teen.

Eye life

Dead eyes are the number-one tell of AI-generated acting. Every character gets continuous, naturalistic ocular life, in the profile and in every scene.

• Micro-saccades and gaze targeting on whatever they attend to; the gaze keeps moving — eyes drift, flick away in thought, scan to a detail and settle back. They never lock frozen.
• Realistic blink rate and quality, tied to state: rapid blink-bursts under stress; slow calm lids in control; a blink-and-glaze on a moment of dissociation.
• Live catchlights — the eyes must read as wet, lit, and alive. But a catchlight is a render property, not a cure [EMPIRICAL — v3.33.0 source drop, re-derived 2026-08-22; source file not recorded]: dead eyes are not fixed by lighting tricks, they are fixed by giving the eyes a task. A glassy stare with a beautiful catchlight is still a glassy stare. Write the eye-work as purposeful action aimed at the partner — checking both of their eyes for a sparkle of trust · registering after each point whether it landed · stealing a look and snapping back before being caught · measuring them, comparing what I feel against what they show. The eye movement IS the doing, and aliveness is the mind visibly working on the task moment to moment. Catchlights make that legible; they never substitute for it.
• Controlled stillness is chosen, never dead. For a near-unblinking predator calm, keep blinks rare, slow, and deliberate — a decision, not a freeze — and let the gaze still shift slowly with intent.
• Eyes lead the thought. The eyes reach the target a touch before the head turns. The thought is readable in the eyes before the words come.
• Eye life reacts to the beat. Blink rate, gaze steadiness, and catchlight warmth all shift when the beat shifts.

The phased-blink shorthand from the film pipeline — one lazy blink → a quick DOUBLE-BLINK → one HARD reset-blink — is the cheapest way to buy a living face in a static shot (../higgsfield-seedance/HELL-GRIND.md § Physics, not adjectives).