name: higgsfield description: > Use this skill whenever the user asks anything about Higgsfield AI — writing or refining video/image prompts, choosing a model (Kling, Sora 2, 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: 3.31.0 updated: 2026-08-09 author: O-Side Media license: MIT

Higgsfield AI Prompt Skill

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

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 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; 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 / 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 (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.

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, Sora 2 incl. Pro/Max/Pro Max tiers (UI-only — confirmed in the UI 2026-07-06, absent from the API/MCP catalog), 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 4.0, GPT Image 2.0, 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.

Working Folders — file handling

The project has a workspace/ folder with three subfolders. Use them for every task that involves a user-provided document or a file you produce. This keeps uploads and deliverables out of the project root.

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."
• These three folders' contents are local-only (git-ignored); do not assume anything in them is committed.

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.

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

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.

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.

Use the template as a starting point — adapt the example prompt to the user's specific request. The annotations explain WHY each element works, helping you make informed substitutions.

Technique templates (templates/seedance/) — structure templates for Seedance prompts where the user request is technique-shaped rather than genre-shaped:

Text-overlay templates (templates/text-overlays/) — paste-ready text-rendering prompts for slogan / subtitle / speech-bubble overlays:

Build the Prompt Using the MCSLA Formula

Full MCSLA definition and prompt structure → skills/higgsfield-prompt/SKILL.md

Quick summary — five layers, every prompt:

Core rules:

• Be specific — name camera presets, describe VFX concretely
• Keep prompts under 200 words (short-form regime — block-scaffold production prompts follow their own structural rules, HARD RULE 8)
• Subject → Action → Camera → Style is the most reliable order

Output Format

Single prompt:

**Model**: [model name]
**Aspect ratio**: [ratio]  **Duration**: [Xs]  **Style**: [style]

[Prompt]

**Camera**: [camera control name]
**Motion preset** (if used): [preset name]

Two versions (when style varies):

### Version 1 — [Style Name]
[Prompt]

---
### Version 2 — [Style Name]
[Prompt]

Output rules:

• Output a clean, ready-to-paste prompt — no meta-commentary after
• Do not explain what every line does unless the user asks
• Always name the camera control and motion preset explicitly

Generation Ledger — log every result

Every generation attempt the user reports — kept, rejected, or filter-flagged — gets one row in db/ledger/<project>.json. The denominator (successes too, not just failures) is what turns the memory system into takes-per-kept ratios and credit budgets.

The 5-second rule: when the user reports a result, ask at most ONE question ("keep or reject — what failed?") and write the row yourself with one scripts/higgsfield_memory.py log-gen command. Never ask twice; never present a form. Full workflow: skills/higgsfield-recall/SKILL.md § Log the Generation Result. Ratios and budgeting: skills/higgsfield-assist/SKILL.md.

@ Reference Rules

• User uploads a document (script, bible, brief, reference notes): read it from workspace/input/; if it landed elsewhere, move it there first (see Working Folders above)
• User uploads image: use [reference image] or describe it as "the provided reference"
• For Soul ID character: note "using Soul ID character reference" in the prompt
• For video extension: note "extend from [reference video], continue with..."
• For style transfer: note "match the visual style of [reference image]"

Shared Resources

Sub-Skills (auto-loaded as needed)

Full vocabulary in vocab.md Full motion preset library in skills/higgsfield-motion/SKILL.md Model comparison in model-guide.md Example prompts in prompt-examples.md Shared negative constraints in skills/shared/negative-constraints.md Genre-specific annotated templates in templates/

Higgsfield AI Prompt Skill

A comprehensive Claude skill library for generating high-quality prompts on Higgsfield AI — the cinematic video and image generation platform.

What This Skill Does

Transforms natural language requests into production-ready Higgsfield prompts using:

• The MCSLA formula (Model · Camera · Subject · Look · Action)
• Named camera controls and motion presets the platform recognizes
• Model selection guidance across Kling 3.0 / 3.0 Omni / 3.0 Motion Control, Sora 2, Veo 3.1, Wan, Seedance 2.5 / 2.0, FLUX 3 Video, Minimax Hailuo, Higgsfield DoP, and more
• Genre recipe templates for action, horror, romance, sci-fi, product ads, and more
• Soul ID character consistency guidance + Character Sheet creation
• Troubleshooting for failed or poor generations
• Cinema Studio 2.5 advanced features: Soul Cast AI actors, built-in color grading, 3D Mode (Gaussian Splatting), Grid Generation, Resolution Settings, Frame Extraction Loop, Object & Person Insertion, Per-Character Emotions, Clustering, Five-View Location Reference Sheet, Reference Sheet Types (Motion / Outfit / Palette), Elements System with library surface (5 source tabs × 6 element categories)
• Cinema Studio 3.0 (Business/Team plan): native dual-channel stereo audio, Smart shot control, 15s max duration, 7 genres, @ reference patterns, Soul Cast 3.0
• Cinema Studio 3.5: three-pill main UI (Genre / Style / Camera), Style Settings panel (8 Color Palette / 6 Lighting / 9 Camera Moveset Style + Manual Style mode), Camera Settings four-axis panel (3 Camera Body / 5 Lens / 5 Focal Length including new 75mm / 3 Aperture), Image Mode with four Cinematic models picker (Soul Cinema default, Cinematic Characters, Cinematic Locations, Cinematic Cameras with 2.5 vocabulary)
• Seedance 2.0 prompting best practices — Intent over Precision, Genre Router, I2V Gate, Anti-Slop, Physics Language, SCELA audio, Reference-Based / Continuation / Expand Shot / Edit Shot / Transformation prompt modes, Continuation Prompt Formula, the Iteration Rule
• Seedance 2.5 omni-reference director — the four generation modes (t2v / omni_reference / video_edit / video_extension), explicit @Image / @Video / @Audio reference roles with exclusions, the five-step multi-reference workflow, 30-second staging with end states, timestamp pacing, bracket syntax for music / SFX / dialogue / subtitles, in-prompt first-last-frame and multi-keyframe control, storyboard grids, coarse-vs-fine blockout rendering, one-click video, seamless transitions, and a 2.0-vs-2.5 routing table
• Hell Grind feature-film pipeline — Higgsfield's open-sourced 95-minute-feature system: headless character sheets, mask-composited point edits, location sheets with anchors and one light logic, the per-scene GEO SPATIAL LAYOUT block, the position-fixing first second, dialogue construction, the ban dictionary, the 10–15 iteration rule, and the crowd / giant / threshold solutions
• AI-VFX pipeline — the production layer under Seedance 2.5, from Higgsfield's AI-vs-VFX challenge build: model routing per asset class (faces / creatures / clothing / locations), the plain-grey sheet law, two-close-up creature sheets, the face-lock crop, the size-ref frame for scale between two subjects, location batching selected on light, the omni_reference video-to-video lane (source ≥4s, duration = source) with its performance-inheritance clause, the four-batch stop rule and the i2v fallback with a deliberate empty-frame stitch point, and a slop catalog of the tells that give a shot away
• Acting director — performance as behavior under pressure: objective / obstacle / tactics / beats / subtext, listening markers, body + status + proxemics, mandatory eye life, the 150–220-word acting master profile and its per-scene rewrite, the locked voice prompt, a 15-symptom atlas of bad acting, and a 0–5 performance scale
• GPT Image 2.0 prompt director — three-format taxonomy (structured JSON for UI mockups / infographics / reference sheets, dense cinematic prose for single-subject scenes, auto-derive meta-prompt for theme-only concepts) plus reference-sheet and static-ad-recreation workflow satellites
• Higgsfield Canvas — node-based / infinite-board workspace guidance: chaining prompts → images → videos, named canvas patterns, Shared Canvas live collaboration, build-free / generate-paid cost model
• Marketing Studio + Content Factory — 9 DTC ad presets (UGC / Tutorial / Unboxing / Hyper Motion / TV Spot / Wild Card / Virtual Try-On) with 4–15s ad video, plus an end-to-end campaign pipeline (research → plan → generate → publish → report) with a cost-savings report
• Shared negative constraints reference — categorized artifacts + prevention phrases (positive alternatives for 3.0); Kling 3.0 Motion Control failure diagnostic; Physics Rendering — Resolution Decision Matrix (cross-model 480p / 720p / 1080p routing rule for Seedance 2.0 + Cinema Studio 3.x)
• Identity vs. Motion separation — hard rule for character consistency across shots
• Annotated templates library — 10 genre templates with Cinema Studio 3.0 genre mappings, plus Seedance technique + character-design + text-overlay sub-libraries (18 files across 3 categories)
• DISCIPLINE.md cross-cutting framework — 9 named discipline patterns in 3-3-3 tier symmetry (prompt-construction, model-selection, iteration-discipline) governing decisions across all sub-skills
• production-benchmarks.md — Hell Grind 90-min Cannes feature reference, per-character iteration anchors, acceptance-rate calibration; what "production quality" means in practice
• FAILURE-MODES.md (Seedance) — 11 named render failures documented with symptom + mechanism + counter for diagnosis-first iteration
• C-arc Building Complete AI Projects — 10-Step Methodology — end-to-end pipeline from idea to delivered project; complements the genre/scene templates
• Expanded Seedance methodology + Soul refinement — Iteration Rule + 6-Pass Diagnostic Sequence + Four Questions + Next-Shot Decision Tree + Bridging / Continuation / Repair working modes; Character Anchor Block + Two-Tool Refinement Pipeline for character consistency at production scale

Install

Claude Code

git clone https://github.com/OSideMedia/higgsfield-ai-prompt-skill ~/.claude/skills/higgsfield

Claude Cowork

Drop the repo folder into your Cowork workspace. The skill dispatcher is at SKILL.md in the repo root.

Claude.ai Projects

Upload SKILL.md (root) as your project instruction base. Upload files from skills/ as project documents.

Higgsfield Stack Integration

This skill is the prompt-construction layer. Higgsfield ships official execution tooling — a CLI, an MCP custom connector, and a bundled skills package. They complement each other: this skill produces the prompt, their tooling executes it. None of their tooling is required for this skill to work — you can always paste prompts directly into higgsfield.ai. But if you want an end-to-end loop, you'll want one of the three.

A Higgsfield account is required for any of the tooling below. Sign up at higgsfield.ai.

Higgsfield CLI

Command-line tool for terminal-native agents (Claude Code, Codex, Cursor). Per Higgsfield's own guidance, prefer the CLI over the MCP if you're working in a terminal.

• Repo: github.com/higgsfield-ai/cli
• Install: curl -fsSL https://raw.githubusercontent.com/higgsfield-ai/cli/main/install.sh | sh or brew install higgsfield-ai/tap/higgsfield
• Auth: higgsfield auth login

Higgsfield MCP

Custom connector for claude.ai web and the Claude desktop app. Separate product from the CLI.

• Connector URL: https://mcp.higgsfield.ai/mcp
• Install: claude.ai → Settings → Connectors → Add custom connector → paste the URL above → sign in

Higgsfield Bundled Skills

Markdown skill bundle for agents that consume Cowork-style skills. All three skills drive the CLI under the hood.

• Repo: github.com/higgsfield-ai/skills
• Install: npx skills add higgsfield-ai/skills
• Skills included: higgsfield-generate, higgsfield-soul, higgsfield-product-photoshoot
• Invoke: /higgsfield:generate, /higgsfield:soul, /higgsfield:product-photoshoot

End-to-end example

How the layers fit together for a real request:

USER:    "Make me a cinematic chase scene through a night market.
          Use my trained Soul character — reference_id abc123."
   ↓
THIS SKILL — higgsfield-ai-prompt-skill
   • routes to higgsfield-prompt + higgsfield-camera + higgsfield-soul
   • picks Kling 3.0 (character-focused, supports --soul-id)
   • applies MCSLA: model, camera preset, subject, look, action
   • appends shared negative constraints
   • outputs a production-grade Higgsfield prompt
   ↓
PRE-FLIGHT (optional, recommended for Veo / Kling / Sora / Seedance video):

   SCHEMA VERIFY (recommended for any model you haven't called recently):
   CLI path:        higgsfield model get kling3_0
                    → returns schema: aspect_ratio enum, duration range,
                      mode/sound options, media roles
   MCP path:        models_explore(action="get", model_id="kling3_0")
                    → returns same schema as CLI

   COST ESTIMATE (no job submitted):
   MCP path:        generate_video(..., get_cost: true)
                    → returns credit cost + adjustme