Imported from osidemedia/higgsfield-ai-prompt-skill (skills/higgsfield-scene-engine/SKILL.md). Install upstream with npx skills add osidemedia/higgsfield-ai-prompt-skill --skill higgsfield-scene-engine. Copyright stays with the author.

Higgsfield Scene Engine — is this scene worth generating?

[DEMO — Tigran (tig-scene-engine), 2026-06-26] [UNPROVEN HERE] These definitions are bespoke and deliberately not the textbook ones. Apply them as written; do not substitute standard screenwriting glosses, which are looser and will pass scenes this engine fails.

Why a prompter carries this. Every shot costs credits and iterations. A structurally dead scene generates just as cleanly as a live one — the model has no opinion about whether a beat earns its place — so the failure surfaces only after the footage exists and the cut does not build. This skill is the cheapest pass in the repo: it runs on text, before any generation, and its whole job is to stop you paying to render a scene that cannot work.

It decides which shots deserve the spend. higgsfield-shotlist-director turns the settled scene into shots; higgsfield-acting writes the performance inside them.

The five-element engine

1. Goal

The single, unchanging thing the hero is fighting for: to fix what has already been established and reach, as fast as possible, the result that resolves it. "Already established" means anything set up earlier — deep backstory, a few scenes ago, or one minute ago.

• The major goal never changes, and the hero is never wrong about it. He can be wrong about tactics, never about the goal.
• Every scene goal must be a causal link toward the story goal. Test: if you can remove the scene and the chain still holds, the scene fails. No orphan scenes.
• "What he ought to do" is not a separate thing to test — the right move is dictated by Obstacle + Stakes. A hero taking a tactic the stakes would not justify (showering while the building burns) is a flaggable inconsistency: name it as bad writing or as deliberate self-sabotage.

2. Obstacle

A strong circumstance that jeopardizes either one stage of the path or the whole goal. It threatens; it does not merely cost.

• It can be a branching search space (many places to look, limited time) or a single hard wall (a dead battery — you cannot search a dead machine). A wall typically spawns a fresh sub-goal with its own terrain.
• Name its scale: local (threatens the current stage) or global (threatens the whole goal).
• The obstacle is what forces the hero to choose or change a tactic. No jeopardy → no tactic needed → no scene.
• Audit test: name what is at risk, and at what scale. If nothing genuinely is, the obstacle is fake and the scene goes slack.

3. Tactic

The move the threat forces the hero to choose — a reasonable guess on his current, often incomplete, knowledge. This is search under uncertainty, not error-as-stupidity.

• A "wrong" tactic is not dumbness; it is a plausible probe of a space he cannot fully see. The engine is a knowledge gap: the distance between what he believes and what is.
• Each failed tactic must return information that narrows the search and reshapes the next move. The drama is the narrowing, not the punishment.
• ✅ good: the tactic is a reasonable bet on current info, and its failure teaches something new.
• ❌ weak: a tactic he had no reason to try, or a failure that returns zero information — a wheel-spin. A failing tactic is only a weakness when the failure teaches nothing.

4. Reversal

A turn against expectation. Three forms:

1. My own action flips on me — what I did for something starts working against it (I shower to make the pitch; the shower makes me late and I miss it).
2. Hidden agency revealed — the thing or person I thought I was acting on was acting on me all along. The thriller move; it often fires exactly when the knowledge gap closes.
3. The general case: the situation turns opposite to what the tactic intended.

Placement: at least one reversal per sequence. A single scene may be pure escalation, but a resolved sequence with zero reversals fails.

A sequence is the run of scenes from when a specific jeopardy opens to when it resolves — overcome, or it defeats him and forces a new path. Sequences nest, and each resolving unit needs its own reversal. Flag which scale you are auditing.

5. Value Shift

The change in the AUDIENCE'S read of a character, triggered by a reversal. The keystone, and the most misunderstood element: it does not happen on the page, it happens in the viewer's mind. Each reversal forces the audience to re-judge the character — a moving verdict they keep revising (nice → experienced → cunning → no, he's desperate → devoted son → what a man).

THE CORE RULE: a reversal with no value shift is inert. A structural turn is not a reversal unless it moves the audience's verdict. If you cannot name a before-verdict and an after-verdict the audience would hold, the turn is dead weight no matter how much plot flipped.

Audit the trajectory, not just presence: are the revaluations building a deepening portrait, or just oscillating?

The causal chain

Goal (fixed; every scene a causal link toward it) → Obstacle (jeopardizes a stage or the whole goal; name what is at risk + scale) → Tactic (forced by the threat; a reasonable guess; its outcome must return information) → Reversal (turn against expectation; ≥1 per resolved sequence) → Value Shift (the audience re-judges the character; no shift → the reversal is useless)

Audit procedure

Work the chain in order. Reversal and Value Shift are judged at the sequence level, not the scene.

1. Goal — state the scene goal in one line. Is it a causal link to the story goal? Apply the removal test. If you do not know the story goal, ask — you cannot fully audit Goal without it.
2. Obstacle — name the circumstance, exactly what it jeopardizes, and the scale.
3. Tactic — is each forced by the jeopardy, reasonable on current knowledge, and does each outcome return information? Flag wheel-spins and unmotivated tactics.
4. Reversal — locate them, name the form, confirm ≥1 per resolved sequence.
5. Value Shift — for each reversal, name before-verdict → after-verdict. If you cannot, mark it INERT and make it the priority fix. Then assess the trajectory.

Audit output

SCENE/SEQUENCE: <one-line identification>

CHAIN CHECK
• Goal — <verdict + one line>
• Obstacle — <what's jeopardized + scale + verdict>
• Tactic — <forced? reasonable? returns info? verdict>
• Reversal — <form + present? ≥1 in sequence? verdict>
• Value Shift — <before-verdict → after-verdict per reversal, or INERT; trajectory note>

WEAKEST POINT: <the single element that, if fixed, recovers the most>

WHAT IF…
1. (Minimal) <change ONLY the weakest point so the rest of their scene still works>
2. (Clean) <a version that fully works, even if it departs further from the original>
3. (Optional) <only if it genuinely adds something>

The three tiers matter in that order: minimal preserves the user's version, clean is yours, and the third is optional. Do not collapse them into one rewrite.

Where this sits before prompting

Once the chain holds, the scene is worth spending on — and only then:

The most common real failure to hunt for first: a reversal that turns the plot but does not move the audience's verdict. Surface it before anything else — it is the one that survives a read-through, generates beautifully, and still lands flat in the cut.

Related Skills

• ../higgsfield-character-design/SKILL.md — premise, world, character, story bible
• ../higgsfield-acting/SKILL.md — the performance layer inside a settled scene
• ../higgsfield-shotlist-director/SKILL.md — scene → shot list