| name | higgsfield-creative-studio |
| description | Orchestrates GPT Image 2.0 still-image prompts and Higgsfield Marketing Studio video-ad prompts into a cohesive AI ad production workflow. Use when the user asks for a full ad asset system, still-to-video campaign package, Higgsfield ad concept, Marketing Studio prompt stack, GPT Image 2 key visual plus video prompt, UGC/video ad prompt pack, product creative system, or wants the Higgsfield and GPT Image 2 prompt directors to work together with marketing, brand, copy, or design strategy. |
Higgsfield Creative Studio
Use this skill as the routing and stacking layer for AI ad assets. Keep the source prompt directors strict:
- For GPT Image 2.0 stills, load
skills/gpt-image-2-director/SKILL.md and obey its output format.
- For Higgsfield Marketing Studio videos, load
skills/marketing-studio-director/SKILL.md and obey its output format.
- If the user asks for a combined system, produce a compact asset package instead of forcing either source skill's single-prompt output format.
- Before any real Higgsfield generation through CLI or MCP, follow
directives/higgsfield-usage-policy.md and run python3 execution/higgsfield_budget_guard.py check. After generation, log with python3 execution/higgsfield_budget_guard.py log.
Credit Guard
Default to prompt-only ideation first. Real generation is budget-gated.
- Prompt-only Creative Capsules are free and do not need user approval.
- Use the guard before any
mcp__higgsfield__generate_image, mcp__higgsfield__generate_video, show_marketing_studio fetch-to-video flow, Soul-ID training, CLI higgsfield generate create, product photoshoot, or marketplace cards command.
- If the guard returns approval required, ask for explicit approval before generating.
- If the guard denies, do not generate; offer a cheaper route such as fewer variants, shorter video, prompt-only package, or next-day run.
- Client work defaults to fewer, stronger outputs. Personal work defaults to quick preview then winner selection.
Router
Choose the path from the user's request:
| Request shape | Load | Output |
|---|
| GPT Image 2 prompt, key visual, poster, landing page mockup, infographic, product frame, storyboard still, character sheet | gpt-image-2-director | Finished GPT Image 2 prompt in the correct source-skill format |
| Higgsfield, Marketing Studio, UGC, Tutorial, Unboxing, Hyper Motion, Product Review, TV Spot, Wild Card, Virtual Try On, video ad prompt | marketing-studio-director | Finished Marketing Studio paragraph plus generation link |
| Photoreal character build, face lock, character sheet, outfit swap, persistent character | skills/banana-pro-director/SKILL.md | Higgsfield still prompt in Banana Pro grammar (18% gray flat plates, Mode 0→1→2A order) |
| Seedance cinematic video prompt | skills/cinema-worldbuilder-pro/SKILL.md | Block-structured Seedance prompt (@tags on Higgsfield surface only) |
| Full ad asset system, still-to-video pack, campaign creative kit, prompt stack | This skill + both source skills as needed | Strategy notes, still prompt, video prompt, QA checklist |
| User asks to actually render/generate/show result | This skill + Higgsfield MCP/CLI bridge | Guarded preview/final generation with logged credit use |
Disambiguation: Higgsfield GPT-2 (face-fidelity image model, credit-heavy — banana-pro-director's escalation path) ≠ OpenAI GPT Image 2 (gpt-image-2-director — layout/typography king, weak faces).
Tool Routing
- Use MCP for in-app widgets, Marketing Studio library/fetch, Soul-ID list/status/training flows, job display, and direct image/video generation when the user benefits from UI.
- Use CLI for local files, account/credit checks,
higgsfield generate cost, product photoshoot backend enhancement, marketplace cards, and repeatable batch workflows.
- For CLI-supported model generation, run
higgsfield generate cost ... first and pass that credit estimate to the guard.
- For product photoshoot, marketplace, Soul-ID, or MCP generation where exact preflight cost is unavailable, use the conservative operation estimates in
.agent/higgsfield-usage.json.
Stacking Order
Use only the stack layers needed for the request.
- Strategy: Load
skills/luke-iha-creative-strategy/SKILL.md for audience, angle, offer, and what-to-say clarity. Load skills/greg-hoffman-brand-mastery/SKILL.md instead when the request is campaign-level, brand-level, or emotional-positioning-led.
- Copy: Load
skills/luke-iha-copy-blocks/SKILL.md when hooks, spoken UGC lines, captions, CTAs, or ad beat language matters.
- Visual Direction: Load
skills/creative-direction/SKILL.md for art direction, mood, platform prompt QA, storyboarding, and visual coherence.
- Prompt Production: Load
gpt-image-2-director for stills and marketing-studio-director for videos.
- Design QA: Load
skills/satori-graphics-design-mastery/SKILL.md for visual communication, layout flow, anti-AI-slop, and message clarity. Use skills/jack-roberts-design-mastery/SKILL.md when the output must align to a reusable design system or multi-format brand package.
Combined Asset Package
When the user asks for both still and video outputs, return:
## Strategy Spine
[1-3 bullets: audience, angle, emotional target]
## Key Visual Prompt
[GPT Image 2.0 prompt. Use a fenced code block and preserve the source skill's JSON/prose/meta-prompt rules.]
## Marketing Studio Prompt
[One flowing paragraph, then blank line, then the Generate link exactly as required by marketing-studio-director.]
## QA Pass
[3-5 checks covering product fidelity, avatar fidelity if applicable, visual hierarchy, brand consistency, and ad clarity.]
## Preview Recommendation
[One guarded first render recommendation: operation, count, duration/resolution if video, and why this is the lowest useful preview.]
Do not add the package wrapper when the user asked only for a single GPT Image 2.0 prompt or a single Marketing Studio prompt. In those cases, follow the source skill's output format exactly.
Image Reference Handling
If product or avatar images are attached:
- Preserve the Marketing Studio fidelity rules: exact product packaging, color, logo placement, proportions, avatar face/build, and user-specified wardrobe.
- Pass image references through with
<<<image_n>>> labels when writing Marketing Studio prompts.
- Do not infer unsupported product claims from images.
- Do not use age markers for avatars.
Quality Gate
Before responding, check:
- The request was routed to the correct source prompt director.
- Any user-specified camera, preset, product, avatar, dialogue, dimensions, or layout requirements survived into the final prompt.
- Combined packages include both still and video prompts only when requested or clearly useful.
- Strict source output formats remain intact for single-prompt requests.
- Any real generation has a guard check and a post-generation log plan.
Execution Prompts (structure-pure v2)
2 deterministic practitioner prompts — each carries an Output Contract, Output Skeleton, and Quality Gate. When a deliverable matches one, Read it and honor its contract instead of improvising the output shape.
- Higgsfield Creative Studio — Combined Asset Package —
skills/higgsfield-creative-studio/references/prompts-v2/combined-asset-package.md
- Higgsfield Creative Studio — Guarded Generation Request —
skills/higgsfield-creative-studio/references/prompts-v2/guarded-generation-request.md
