higgsfield — скилл для Claude Code: что делает и как установить

---
name: higgsfield
description: >
  Writes and refines prompts for Higgsfield AI's video and image models, and picks the
  model, camera controls and motion presets for each shot. Use whenever the user asks
  anything about Higgsfield AI — writing or
  refining video/image prompts, choosing a model (Kling, Veo, Wan, Seedance,
  Minimax Hailuo, DoP, Soul, Nano Banana, Seedream, Flux, GPT Image, etc.), camera
  controls, named motion presets, Soul ID character consistency, Cinema Studio 2.5/3.0,
  Vibe Motion, troubleshooting failed generations, credit optimization, Photodump,
  or any mention of higgsfield.ai. Also trigger on generic "write me a video prompt"
  or "make me an AI video prompt" requests when Higgsfield is the user's configured
  platform.
user-invocable: true
metadata:
  tags: [higgsfield, video, image, prompt, cinematic, AI, filmmaking, motion, camera]
  version: 3.42.0
  updated: 2026-10-23
  author: O-Side Media
  license: MIT
---

# Higgsfield AI Prompt Skill

<!-- CONTENTS — generated by scripts/build_index.py; do not hand-edit -->
**Contents**

- [HARD RULES — pre-delivery checklist](#hard-rules--pre-delivery-checklist)
- [What Is Higgsfield?](#what-is-higgsfield)
- [Working Folders — file handling](#working-folders--file-handling)
- [Handling a Request](#handling-a-request)
  - [Fast Path — Simple Creative Requests](#fast-path--simple-creative-requests)
  - [Full Path — Production Requests](#full-path--production-requests)
  - [Route to the Right Skill](#route-to-the-right-skill)
  - [Load Map — how much to read](#load-map--how-much-to-read)
  - [Check Templates for Genre Match](#check-templates-for-genre-match)
  - [Build the Prompt Using the MCSLA Formula](#build-the-prompt-using-the-mcsla-formula)
  - [Output Format](#output-format)
- [Generation Ledger — log every result](#generation-ledger--log-every-result)
- [@ Reference Rules](#-reference-rules)
- [Shared Resources](#shared-resources)
  - [Reference files inside sub-skills](#reference-files-inside-sub-skills)
- [Sub-Skills — open `skills/<name>/SKILL.md` yourself](#sub-skills--open-skillsnameskillmd-yourself)

<!-- /CONTENTS -->

**Language rule:** Reply in whatever language the user writes in.

**Paths:** every path in this file (`skills/…`, `templates/…`, `specs/…`, `scripts/…`,
`db/…`) is relative to the folder this `SKILL.md` lives in, not to the user's working
directory. A path inside any other file is relative to that file's own folder
(`../../specs/…` in a sub-skill climbs back to this one). Two exceptions:
`workspace/` lives in the user's working directory, and the learning memory may live
outside this folder (both in § Working Folders). Sub-skills are plain
files, not separately registered skills: nothing loads them for you, so open
`skills/<name>/SKILL.md` yourself.

**MCP tools:** `Higgsfield:<tool>` (e.g. `Higgsfield:models_explore`) means that tool on
the Higgsfield MCP server (`https://mcp.higgsfield.ai/mcp`). Its name in your tool list
depends on how it was added — the claude.ai connector shows up in Claude Code as
`mcp__claude_ai_Higgsfield__<tool>` — so match on the Higgsfield server, never on a
same-named tool from another server. A bare tool name in a sub-skill means the same.

---

## HARD RULES — pre-delivery checklist

These rules apply to every Higgsfield response. They are written as a pre-delivery checklist the agent runs *before* sending the response, not as prohibitions stated and then forgotten. The failure mode they prevent is **plausibility-over-verification** — producing a response that looks correct because the agent's training data knows the rough shape of Higgsfield work, rather than because the agent actually read the skill files and verified the platform's ground truth.

**Before delivering any Higgsfield response, confirm in this order:**

1. **Routing line present.** First line of response names which sub-skills you routed to (e.g. "Routing to higgsfield-prompt + higgsfield-camera for an Atmosphere push-in"). One line, then the work. Missing routing line = response is incomplete; add it.

2. **Routed sub-skills opened and read in this conversation.** Match the user's ask to the routing table below, open the matching sub-skill files (`skills/<name>/SKILL.md`) with the read tool, and READ them. Root `SKILL.md` and `skills/higgsfield-prompt/SKILL.md` are mandatory at minimum on any prompt request. Grepped snippets do not satisfy this rule. Full reads do. If your only access to root `SKILL.md` or `skills/higgsfield-prompt/SKILL.md` in this conversation came from grep results, you have not satisfied this rule — open the file. Platform vocabulary, preset names, and model parameters must come from the files because this platform's lineup changes between releases.

3. **Named vocabulary verified, not invented.** Camera preset names, motion preset names, model names, CLI flag forms, and MCP tool parameter names all come from the skill files or from verification. For model parameters, enums, and durations, verify against `specs/model-specs.yaml` first — it is generated from a dated `models_explore` snapshot (see `snapshot_date` inside the file); if the snapshot is stale (>30 days), verify live instead (`higgsfield model get <model>` for CLI param schemas; `Higgsfield:models_explore` for MCP). If you found yourself thinking "this flag probably looks like X" or "this preset is probably called Y" — stop. Read the file or run the verification command. Plausibility is not validity. Do not substitute generic video-prompt vocabulary for named Higgsfield presets; do not invent model versions, camera presets, or motion-preset names. If the user names one you don't see in the skill files, say so and ask for clarification.

4. **MCSLA structure intact on video prompts.** Model · Camera · Subject · Look · Action. Five layers, every video prompt, unless the user explicitly opted out.

5. **Shared negative constraints appended.** Pull positive-phrasing prevention phrases from `skills/shared/negative-constraints.md`. Do not paraphrase from training; use the exact phrasing from the file. (Kling 3.0 prefers positive phrasing over negations; using negation-form constraints when the file says positive is a fidelity miss.)

6. **Preflight surfaced when applicable.** If execution intent is signaled (CLI / MCP / bundled-skills mentioned) AND a video-class or high-cost model is named OR a budget concern is named, surface the two-step preflight (`model get` / `Higgsfield:models_explore` for schema, then cost estimate). See `skills/higgsfield-stack/SKILL.md` § Preflight discipline.

7. **Aspect ratio is an enum, not a free-form value.** Check the model's allowed ratios against `specs/model-specs.yaml` before writing them into the header; if the snapshot is stale (>30 days), verify live via schema (`Higgsfield:models_explore` / `model get`). Example of why this matters: Seedance 2.0 supports native 21:9, Kling 3.0 does not. Anamorphic / 2.35:1 / 2.39:1 are *style register* vocabulary for the Look line, not output ratios. See `vocab.md` § Aspect Ratio: output spec vs. style register. In a multi-shot Seedance sequence whose location plates carry a baked lens, whether the optics words still belong in the video prompt (Style Prefix included) is OPEN — one studio drops them, Hell Grind keeps the look in both (`skills/higgsfield-seedance/SKILL.md` § Bake it into the asset; `skills/shared/house-rulings.md` P2-6).

8. **Prompt under 200 words — short-form regime only.** Soft cap from MCSLA section. Going over is a signal you're padding rather than locking — tighten. **Regime exception:** block-scaffold production prompts (`skills/higgsfield-seedance/SKILL.md` § Official Prompt Architecture) replace the word cap with structural lint — harvested production Seedance briefs run 218–2,059-word medians depending on register `[FIELD — 13-project community harvest, 2026-07-18]`. The cap governs single-shot MCSLA prompts; a block-scaffold prompt over 200 words is not a rule-8 violation.

**If any of items 1–8 are missing or unverified, the response is incomplete. Complete them before sending, not after.**

---

## What Is Higgsfield?

Higgsfield is a cinematic AI video and image generation platform built for filmmakers and
creators. Unlike single-model tools, Higgsfield hosts **multiple generation engines** on one
platform — Kling 3.0/3.0 Omni/3.0 Motion Control, Google Veo 3.1/3.1 Lite, Wan 2.7/2.6/2.5,
Seedance 2.5/2.0/Pro, FLUX 3 Video, Minimax Hailuo 2.3/02, Higgsfield DoP (Lite/Standard/Turbo) for video; Soul 2.0, Soul Cinema Preview,
Soul Cast, Nano Banana Pro/2, Kling Image 3.0/Omni, Seedream 5.0 Pro/Lite/Flash + 4.5, GPT Image 2.0 / 2.5,
Flux 2/Kontext for images — plus a library of 100+ named **Motion Presets**, a **Soul ID**
character consistency system, **Cinema Studio 2.5**, **Cinema Studio 3.0** (Business/Team plan), and **Cinema Studio 3.5** with Soul Cast AI actors, native dual-channel stereo audio, and 80+
one-click **Apps**. **Sora 2 is retired from this skill's recommendations:** OpenAI shut the Sora 2 API down on 2026-09-24; Higgsfield only ever offered it in its web UI, and whether the UI still does is unconfirmed. For scale / physics shots use Seedance 2.0 or Minimax Hailuo 2.3 (`model-guide.md`). **Catalog check (2026-09-26):** Kling 3.0 Omni, Wan 2.5, Seedance Pro, Higgsfield DoP, Soul Cinema Preview and Kling Image 3.0 / Omni are not in the API catalog (`higgsfield model list` / `workflow list`). They may be UI-only — verify in the live UI before recommending them, and prefer a catalog pick where `model-guide.md` names one. No catalog variant of Minimax Hailuo is named 02.

---

## Working Folders — file handling

`workspace/` lives in the **user's current working directory** (the folder the
session was started in), not in this skill's folder. Use its three subfolders for
every task that involves a user-provided document or a file you produce. This keeps
uploads and deliverables out of the project root.

- **Session started inside this repo** (Cowork with the repo folder as the workspace,
  or Claude Code opened here): it is this repo's own `workspace/`.
- **Skill installed elsewhere** (e.g. `~/.claude/skills/higgsfield/`) **while working in
  another project:** use `<that project>/workspace/`, creating `input/`, `output/` and
  `processed/` there on first use. Never put the user's files in the skill's install
  folder — a reinstall replaces that folder and everything in it.

| Folder | Role | Your behavior |
|--------|------|---------------|
| `workspace/input/` | Documents the user wants you to read (scripts, story bibles, briefs, character sheets, references). | Read from here first. If the user's file is somewhere else, **copy it into `workspace/input/`** and leave the original where it is — never move or delete it. When you need a document from the user, ask them to drop it in `workspace/input/`. |
| `workspace/output/` | Files you generate for the user (prompt packs, shot breakdowns, batch CSVs, reports, exported docs). | **Write every file deliverable here**, not to the project root. Tell the user the path when you finish. |
| `workspace/processed/` | Inputs you have finished consuming. | When a task is complete, **move the file from `input/` to `processed/`** so `input/` stays clean. Never delete the user's files — relocate them. |

Rules:
- Never scatter user uploads or generated files across the project root or skill folders — route them through `workspace/`.
- Treat `workspace/input/` as the canonical place to look when the user says "the script / bible / reference I gave you."
- In this repo the three folders' contents are git-ignored. In another project they are not unless the user ignores them: never assume anything in them is committed or private, and ask before editing the user's `.gitignore`.

**The memory folder.** The filter memory, quality memory and generation ledger live in
one folder, and only `scripts/higgsfield_memory.py` reads or writes it. When this skill
runs from a Claude Code skill install (a `.claude/skills/<name>/` folder), the memory
folder is `~/.higgsfield-skill/db/`, copied from this skill's `db/` on first use, so a
reinstall or update cannot erase it. Anywhere else (a Cowork workspace, a clone of the
repo) it is this skill's own `db/`. The `HF_DB_DIR` environment variable overrides both.
`python3 <this skill's folder>/scripts/higgsfield_memory.py where` prints the folder in
use; give the user that path when they ask where their memory is or before a reinstall.
If its `unmerged` list is not empty, another install held entries the memory folder
lacks (its own memory from before v3.42.0, or starter entries from a newer release): they
were kept there, not merged. Tell the user once.

## Handling a Request

### Fast Path — Simple Creative Requests

If the user provides a clear creative intent ("write me a prompt for a car chase at night")
with no specific constraints, **generate immediately** using these sensible defaults:

> **Fast Path still requires reading `skills/higgsfield-prompt/SKILL.md` first — Fast Path means skip clarifying questions, NOT skip the file read.**

| Parameter | Default |
|-----------|---------|
| Aspect ratio | 16:9 |
| Duration | 8s (Kling lanes — see Seedance exception below) |
| Style | Cinematic |
| Video model | Kling 3.0 (character-focused) or Seedance 2.0 (action/scale/references) |
| Image model | Soul 2.0 (portrait) or Nano Banana 2 (everything else) |

Do not ask clarifying questions. Deliver a ready-to-paste prompt. Mention the defaults
used so the user can adjust if they want something different.

> **Seedance exception:** Seedance 2.0 never gets a silently defaulted runtime
> (`skills/higgsfield-seedance/SKILL.md` — always ask, never default). On Fast
> Path that means: if the user named no duration, route video to Kling 3.0;
> pick Seedance 2.0 only when the request names a duration — or when the user
> asked for Seedance by name, in which case state the assumed runtime as the
> first adjustable default in the delivery.

> If you did not read `skills/higgsfield-prompt/SKILL.md` earlier in this conversation, read it now before writing the prompt.

### Full Path — Production Requests

When the user signals production-grade intent (Cinema Studio, multi-shot, specific model,
budget constraints, client work), **confirm before generating:**

**Required:**
- **Generation type**: Image / Video / App (one-click)
- **Video duration**: model-dependent enum — check the model's `duration` values in `specs/model-specs.yaml` before offering choices (e.g. Seedance 2.0 4–15s, Veo 3.1 4/6/8s; image-to-video clips trend short)
- **Aspect ratio**: 16:9 / 9:16 / 1:1 / 4:5 / 4:3 / 21:9-where-supported (default: 16:9) — enums per model in `specs/model-specs.yaml` (HARD RULE 7); anamorphic / 2.35:1 / 2.39:1 are **Look-line style register**, never output ratios
- **Model preference** (or ask Claude to recommend — see `skills/higgsfield-models/SKILL.md`)

**Optional (skip if user already provided):**
- Visual style: Cinematic / VHS / Super 8MM / Anamorphic / Abstract
- Soul ID character reference (if character consistency needed)
- Reference image for image-to-video
- Motion preset preference

> Ask everything in one message — do not split across multiple rounds.

---

### Route to the Right Skill

| User wants | Route to |
|------------|----------|
| User unsure which workspace/tool fits, or asks "what should I use for X" | `skills/higgsfield-workspaces/SKILL.md` |
| Write or improve a prompt | `skills/higgsfield-prompt/SKILL.md` + relevant sub-skills |
| Develop a character / world / story / premise before prompting, build a character sheet / story bible, lock a visual style ("visual DNA"), keep a character consistent across many shots, or "I keep getting generic AI characters" | `skills/higgsfield-character-design/SKILL.md` |
| Audit or strengthen a scene / sequence / beat outline before generating it, or "is this scene working", "what's weak here", "why doesn't this land" | `skills/higgsfield-scene-engine/SKILL.md` |
| Cinematic still image prompt (shot framing, angles) | `skills/higgsfield-image-shots/SKILL.md` |
| GPT Image 2.0 / 2.5 / gpt-image-2 / gpt_image_2_5 prompt, Flare / Sunburst, transparent-background image, UI mockup, infographic, character/reference sheet, layout-dense image, or static-ad recreation | `skills/higgsfield-gpt-image-2/SKILL.md` |
| Choose the right model — incl. which lane edits existing footage, which model makes one clip longer than 15s, or motion transfer (Genjutsu vs Kling Motion Control) | `skills/higgsfield-models/SKILL.md` + `model-guide.md` (§ Edit-Lane Chooser, § Long-Take Chooser) |
| Camera movement guidance (video) | `skills/higgsfield-camera/SKILL.md` |
| Named motion preset (Explosion, Werewolf, etc.) | `skills/higgsfield-motion/SKILL.md` |
| Visual style selection | `skills/higgsfield-style/SKILL.md` |
| Character consistency across shots | `skills/higgsfield-soul/SKILL.md` |
| **Consistency tie-break:** character consistency via a live Soul ID / reference images inside Higgsfield → `skills/higgsfield-soul/SKILL.md`; developing WHO the character is first (sheet, story bible, visual DNA) → `skills/higgsfield-character-design/SKILL.md`. Both may apply in sequence: design first, then lock with Soul. | — |
| VFX presets (Air Bending, Plasma, etc.) | `skills/higgsfield-motion/SKILL.md` |
| One-click App workflow | `skills/higgsfield-apps/SKILL.md` |
| Genre recipe (action, horror, ad, etc.) | `skills/higgsfield-recipes/SKILL.md` |
| Fix a failing generation | `skills/higgsfield-troubleshoot/SKILL.md` |
| A Seedance take came back wrong (not flagged) — reversal, babble, a third hand, gliding walk, choppy fight | `skills/higgsfield-seedance/SKILL.md` (`skills/higgsfield-seedance/FAILURE-MODES.md`) + `skills/higgsfield-troubleshoot/SKILL.md` (§ Stop-Rule Ladder) |
| Moodboard, style direction, Soul Hex color | `skills/higgsfield-moodboard/SKILL.md` |
| Visual consistency across a project | `skills/higgsfield-moodboard/SKILL.md` |
| Mixed Media presets (Noir, Sketch, Particles, etc.) | `skills/higgsfield-mixed-media/SKILL.md` |
| Photodump style preset / social-feed photo-dump aesthetic | `photodump-presets.md` (root reference) |
| Artistic style transformation, preset stacking | `skills/higgsfield-mixed-media/SKILL.md` |
| Higgsfield Assist (GPT-5 copilot) | `skills/higgsfield-assist/SKILL.md` |
| Credit optimization, plan selection, budget strategy | `skills/higgsfield-assist/SKILL.md` |
| Cinema Studio 2.5 / Cinema Studio 3.0 / Cinema Studio 3.5 / Cinema Studio 4.0 / multi-shot sequence workflow / Soul Cast | `skills/higgsfield-cinema/SKILL.md` |
| Cinema Studio 4.0 — `cinematic_studio_video_4_0`, its four modes (t2v / omni_reference / video_edit / video_extension), camera / lens / aperture / era / genre / pacing ids, light and palette controls | `skills/higgsfield-cinema/SKILL.md` |
| Optical physics, camera bodies, lenses, Hero Frame | `skills/higgsfield-cinema/SKILL.md` |
| Elements system (@Characters/@Locations/@Props) | `skills/higgsfield-cinema/SKILL.md` |
| Director Panel, Speed Ramp, shot modes, Popcorn | `skills/higgsfield-cinema/SKILL.md` |
| Cinema Studio 3.0 Smart mode, @ references, native audio | `skills/higgsfield-cinema/SKILL.md` |
| Cinema Studio 3.5 — three-pill UI, Style Settings, Camera Settings, Manual Style, AI director toggle | `skills/higgsfield-cinema/SKILL.md` |
| User mentions Marketing Studio, DTC Ads, `ms_image`, or `marketing_studio_video` model | `skills/higgsfield-marketing-studio/SKILL.md` |
| User wants UGC / Tutorial / Unboxing / Hyper Motion / Product Review / TV Spot / Wild Card / UGC Virtual Try On / Pro Virtual Try On ad video | `skills/higgsfield-marketing-studio/SKILL.md` |
| User mentions hook+setting picklists, preset / custom / text-generated avatars in MS context, or 4–15s ad video constraints | `skills/higgsfield-marketing-studio/SKILL.md` |
| **Ad Multiplier** ("multiply my ad", many independently edited versions of one 4–30s ad) or a one-off **Genjutsu** edit of a finished ad (swap one object / product / garment using a reference image of the new one, transfer motion) | `skills/higgsfield-marketing-studio/SKILL.md` (§ 14) — a swap described in words only, with no reference image of the replacement, is Seedance 2.5 `video_edit` (`model-guide.md` § Edit-Lane Chooser) |
| User wants to run a full campaign pipeline — research → plan → generate → publish → report, "create a campaign", "100 UGC videos", content plan, batch ads, cost-savings report | `skills/higgsfield-content-factory/SKILL.md` |
| User mentions Higgsfield Canvas, a node-based / node-graph workspace, an infinite board, chaining prompts→images→videos into a pipeline, Shared Canvas, or a ComfyUI-style node workflow | `skills/higgsfield-canvas/SKILL.md` |
| Multi-shot workflow, chaining tools, full production pipeline | `skills/higgsfield-pipeline/SKILL.md` |
| Short film, branded content, Popcorn → video → assembly | `skills/higgsfield-pipeline/SKILL.md` |
| Vibe Motion, kinetic typography, animated text, infographic/data/presentation animation, logo/brand animation as **code** (Remotion — crisp text, exact brand colors, deployable, real-time edits) | `skills/higgsfield-vibe-motion/SKILL.md` |
| Animated AD / brand promo as an **AI-generated video** built brief → storyboard sheet → Seedance ("make a motion", "motion design ad", "animate my logo into a video", "promo/ad video", classicMD/highMD) | `skills/higgsfield-motion-design/SKILL.md` |
| **Vibe Motion vs Motion Design** (both say "motion graphics / brand / logo animation"): want **crisp text, exact colors, deployable code, editable canvas** → `skills/higgsfield-vibe-motion/SKILL.md` (Remotion code); want a **cinematic/kinetic AI video clip** (pixel render, native audio, no guaranteed-crisp text) → `skills/higgsfield-motion-design/SKILL.md` (Seedance). When unsure, ask "should the text/logo stay perfectly crisp and editable (code), or is this a rendered video clip?" | — |
| Pre-generation memory check, apply past failure fixes | `skills/higgsfield-recall/SKILL.md` |
| User reports a generation result (kept/rejected/flagged) — log it to the ledger | `skills/higgsfield-recall/SKILL.md` |
| Takes-per-kept ratios, credit budgeting from logged data | `skills/higgsfield-assist/SKILL.md` |
| Audio design, dialogue cues, SFX, ambient sound | `skills/higgsfield-audio/SKILL.md` |
| **Standalone audio generation** — soundtrack, ambience bed, multi-speaker scene audio, Seed Audio 1.0 (`seed_audio`), TTS voiceover / narration as its own deliverable | `skills/higgsfield-audio/SKILL.md` |
| Swap or revoice the speaker in a finished video (`Higgsfield:voice_change`), or clone / create a reusable voice (`Higgsfield:create_voice` → `voice_type: element`) | `skills/higgsfield-audio/SKILL.md` |
| **A finished video in, something derived out** — Shorts Studio (restyle one video into shorts), Clipify (YouTube → subtitled clips), Virality Predictor ("will this go viral", score my ad), Video Analysis (scene-by-scene breakdown), and which of these cost credits | `skills/higgsfield-repurpose/SKILL.md` |
| **3D** — image→3D / multi-view→3D / text→3D mesh (GLB), rigging or animating a mesh, remesh / retexture, 3D Body, 3D Jutsu / Scene Builder 3D projects, a 3D turnaround as a multi-angle reference, a 3D blockout as a staging reference (`Higgsfield:generate_3d`, `Higgsfield:scene_builder_3d_*`) | `skills/higgsfield-3d/SKILL.md` |
| **Extend / continue an existing clip** — "make it longer", "what happens next / before", prequel, last-frame handoff, extension chains | a native forward/backward extension on 2.5 → `skills/higgsfield-seedance-2-5/SKILL.md` (`video_extension`); a 2.0 clip continued as a video reference → `skills/higgsfield-seedance/SKILL.md` (§ Extension Prompting); chaining across generations → `skills/higgsfield-pipeline/SKILL.md` (§ Continuation & Extension Handoff) |
| **Prep assets / reference sheets before video** — character sheet, prop three-view, location plate, "build my elements", variety sheet for crowds | `templates/ad-asset-prep.md` + `skills/higgsfield-gpt-image-2/SKILL.md` (props) + `skills/higgsfield-soul/SKILL.md` (people & crowds) |
| Audition / screen-test a designed character (how they move, speak, react) before scene generation | `skills/higgsfield-character-design/SKILL.md` (§ Screen Test / Audition) |
| Seedance 2.0 / Pro prompt, flagged prompt, credit waste on Seedance | `skills/higgsfield-seedance/SKILL.md` |
| **Seedance 2.5** — user names 2.5 / Dreamina / Jimeng, wants a single clip longer than 15s, wants to **edit** or **extend** a video that already exists, or supplies many image/video/audio references (up to 30/10/10) | `skills/higgsfield-seedance-2-5/SKILL.md` |
| **2.0 vs 2.5** (both say "Seedance"): needs 4K, `mode=fast`, or a `genre` hint → `skills/higgsfield-seedance/SKILL.md`; needs >15s in one generation, video editing, forward/backward extension, or heavy multi-reference → `skills/higgsfield-seedance-2-5/SKILL.md`. Both do 1080p and platform start/end frames (2.5 only in `omni_reference`); 2.5 caps at 1080p | — |
| **Character performance** — acting, behavior, mannerisms, tics, a gait, subtext, "my characters look wooden / dead-eyed / AI", keeping a character themselves across many shots, an acting master profile | `skills/higgsfield-acting/SKILL.md` |
| **Feature-film production pipeline on Seedance** — headless character sheets, location sheets, a scene geography block reused across shots, dialogue construction, iteration discipline, giants / crowds / threshold transitions | `skills/higgsfield-seedance/SKILL.md` (`skills/higgsfield-seedance/HELL-GRIND.md`) |
| **Transform footage the user already has** (video-to-video): "make a Seedance prompt for this video/clip", add a VFX element (set my head/hair on fire, transform my hand, make a limb invisible), swap the world/background around a preserved subject (desert, clouds, lava, neon city), put a giant creature behind me or on a landmark, relight/regrade to match, sync a crash-zoom/push-in to a line — a **real source clip** is the starting point | `skills/higgsfield-seedance-vfx/SKILL.md` |
| **Replace a VFX/3D pipeline with generation** — "can AI do this instead of VFX", put myself in this plate, put a creature in my footage, a dragon/monster shot without buying or rigging a 3D model, "how do I build a whole VFX shot", asset sheets → size-ref → locations → shots as one pipeline; also "my v2v keeps failing / turns to slop", scale drift between two subjects, which image model for faces vs creatures vs clothing vs locations | `skills/higgsfield-seedance-2-5/SKILL.md` (`skills/higgsfield-seedance-2-5/VFX-PIPELINE.md`) |
| Precise facial expression / FACS / Action Unit codes (AU12, AU6…), forced or uncanny or mixed expression, close-up micro-performance, monologue/dialogue facial acting, "which AU code for anger/fear", FACS reference sheet | `skills/higgsfield-facs/SKILL.md` |
| "Make a shotlist", break a script/brief/treatment into many connected Seedance prompts, director's shotlist, global style prefix + `@`-glossary + named per-scene prompts as one editable HTML | `skills/higgsfield-shotlist-director/SKILL.md` |
| User has Higgsfield CLI / MCP / bundled skills installed and asks how this skill works alongside them | `skills/higgsfield-stack/SKILL.md` |
| User mentions `higgsfield auth login`, `higgsfield generate create`, `mcp.higgsfield.ai/mcp`, `/higgsfield:generate`, or asks "do I need both" | `skills/higgsfield-stack/SKILL.md` |
| Request is squarely one of Higgsfield's own connector workflows (character-sheet, thumbnail-generation, ugc-*, faceless-video, narrator, subtitles, product-photoshoot, brand-asset-creation, ad-multiplier, video-editing / Higgsedit) — what this library adds vs hands off | `skills/higgsfield-stack/SKILL.md` |
| User asks where the prompt construction ends and the CLI/MCP execution begins (handoff questions) | `skills/higgsfield-stack/SKILL.md` |

---

### Load Map — how much to read

The routing table says *where*; this says *how much*. Loads are cumulative — every path starts from HARD RULE 2's mandatory reads.

| Situation | Load |
|-----------|------|
| Simple creative prompt (Fast Path) | root `SKILL.md` + `skills/higgsfield-prompt/SKILL.md` — nothing else |
| Any Seedance prompt | + `skills/higgsfield-seedance/SKILL.md` (+ the matching `templates/seedance/` file when the request is technique-shaped) |
| Multi-scene / sequence / script breakdown | + `skills/higgsfield-shotlist-director/SKILL.md` + `skills/higgsfield-pipeline/SKILL.md` |
| Model choice unclear or contested | + `skills/higgsfield-models/SKILL.md` + `specs/` (the generated spec for the output type) |
| User reports a generation result | + `skills/higgsfield-recall/SKILL.md` (ledger write) |
| A Seedance render failed (not filtered) | + `skills/higgsfield-seedance/FAILURE-MODES.md` + `skills/higgsfield-troubleshoot/SKILL.md` |
| Budget / credits / plan question | + `skills/higgsfield-assist/SKILL.md` |
| Anything else | one routing-table row → that sub-skill; resist loading more than the row names |

### Check Templates for Genre Match

Before writing a prompt from scratch, check if the user's request matches a common genre
pattern. The `templates/` folder contains 10 annotated example templates with line-by-line
breakdowns, recommended models, negative constraints, and variations.

| User request matches | Check template |
|---------------------|----------------|
| Chase, pursuit, action, parkour | `templates/01-cinematic-action-chase.md` |
| Product, commercial, ad, UGC | `templates/02-product-ugc-showcase.md` |
| Horror, scary, creepy, dread | `templates/03-horror-atmosphere.md` |
| Fashion, editorial, lookbook | `templates/04-fashion-editorial.md` |
| Sci-fi, cyberpunk, VFX, space | `templates/05-sci-fi-vfx.md` |
| Portrait, character intro, close-up | `templates/06-portrait-character-intro.md` |
| Landscape, natu...