skills README

项目的 README,来自 github.com/higgsfield-ai/skills。合并为一个 markdown 文件:README.md。

Higgsfield AI Skills

License: MIT Version Skills Discord

AI agent skills for image/video generation and complete visual brand systems via Higgsfield AI, including Marketing Studio and Virality Predictor video scoring. Works with Claude Code, Cursor, Codex, and other AI coding agents that load Markdown-based skills.

Install

Pick one. Each method handles the Higgsfield CLI install and auth as part of skill setup.

npx skills add higgsfield-ai/skills

gh skill install

GitHub CLI v2.90+:

gh skill install higgsfield-ai/skills

Claude Code marketplace

Inside Claude Code:

/plugin marketplace add higgsfield-ai/skills
/plugin install higgsfield@higgsfield

Setup script

Universal fallback:

git clone --depth 1 https://github.com/higgsfield-ai/skills.git
cd skills
./setup

More options in INSTALL.md. Agent-driven install (paste into your agent): INSTALL_FOR_AGENTS.md.

Skills

Skill Invoke (Claude Code) Description
higgsfield-generate /higgsfield:higgsfield-generate Image, video, 3D, and audio generation across 30+ models (Nano Banana 2, Soul V2, Veo 3.1, Kling 3.0, Seedance 2.5, Seed Audio 1.0, Flux 2, GPT Image 2.6 with native GIF/SVG, …), plus Marketing Studio for branded ads with avatars/products/hooks/settings and Virality Predictor scoring for finished videos.
higgsfield-soul-id /higgsfield:higgsfield-soul-id Train a Soul Character — a reusable, face-faithful identity model. Returns a reference_id consumable by Soul-aware generation.
higgsfield-product-photoshoot /higgsfield:higgsfield-product-photoshoot Brand-quality product imagery with mode-specific prompt enhancement. 10 modes (studio, lifestyle, Pinterest, hero banner, ad packs, virtual try-on, …) backed by gpt_image_2.
higgsfield-brandkit /higgsfield:higgsfield-brandkit Create or extend a complete visual identity: palettes, editable SVG logo marks, typography, mockups, social graphics, packaging, signage, merchandise, decks, and canonical PPTX/PDF brandbooks.
higgsfield-marketplace-cards /higgsfield:higgsfield-marketplace-cards Marketplace product cards: compliant main image, secondary product images, and A+ style modules via backend prompt enhancement.
higgsfield-websites /higgsfield:higgsfield-websites Build, edit, and deploy full-stack websites — React 19 + TanStack Start, server-rendered as one Cloudflare Worker with D1 / R2 / KV / Durable Objects / Containers. Create → get git repo access → edit locally → push → deploy preview/production, all via higgsfield website ….
higgsfield-faceless-video /higgsfield:higgsfield-faceless-video Create faceless explainers, stories, documentaries, and children's music videos.
higgsfield-video-explainer /higgsfield:higgsfield-video-explainer Compatibility entry for faceless explainers.
higgsfield-youtube-thumbnail /higgsfield:higgsfield-youtube-thumbnail Create truthful, high-impact YouTube thumbnails and vertical video covers with identity-preserving references, controlled variants, focused edits, and optional text-overlay guidance.

The skills chain: train Soul → use the reference id in generate (including Marketing Studio jobs). brandkit chains Recraft, Seedream, GPT Image, and deterministic local tooling into one approval-aware identity system. product-photoshoot and marketplace-cards are self-contained — backend enhances prompts before submitting image jobs. youtube-thumbnail can follow any video workflow once its truthful topic and visual direction are known. websites chains with generate to embed bespoke hero images, video loops, and OG assets in the site.

Modes

higgsfield-product-photoshoot — 10 modes for brand visuals:

Mode What it's for
product_shot Product on neutral / studio / catalog background
lifestyle_scene Product in a real environment — hands, action, atmosphere
closeup_product_with_person Tight crop with hands or partial face — beauty, demonstrating
moodboard_pin Vertical 2:3 Pinterest-native pin, moodboard feel
hero_banner Wide-format website / email / campaign header
social_carousel 3–10 connected slides for IG / LinkedIn / Facebook
ad_creative_pack Coordinated pack of static ad variants for Meta / TikTok / Pinterest / Google
virtual_model_tryout Product worn or used by an AI-rendered model
conceptual_product Surreal / CGI-style / levitating / splash / sculptural product
restyle Transform an existing image's aesthetic, mood, or seasonal context

higgsfield-generate Marketing Studio — 9 modes for branded ad video:

Mode What it's for
ugc Default. Casual, organic-feel content from a presenter
ugc_how_to Tutorial / explainer
ugc_unboxing Unboxing reveal
product_showcase Clean product highlight, polished
product_review Presenter giving an opinion
tv_spot Broadcast-style commercial
wild_card Experimental, model picks the vibe
ugc_virtual_try_on Trying on clothing — UGC vibe
virtual_try_on Trying on clothing — polished, model-driven

Quick Reference

What you want Skill Note
Generate any image / video from a prompt higgsfield-generate Prefers gpt_image_2_6 / nano_banana_flash for images and SOTA seedance_2_5 for video by default
Generate audio from a prompt higgsfield-generate Prefers seed_audio by default
Image with my own face higgsfield-soul-id then higgsfield-generate One-time training, then --soul-id
Branded product photo (studio / lifestyle / Pinterest / hero / ad pack) higgsfield-product-photoshoot Mode-specific prompt enhancer + gpt_image_2
Create or extend a logo, visual identity, brand asset system, or Brandbook higgsfield-brandkit Recraft SVG marks + approved palette/type + local editable exports and dependency-aware revisions
Marketplace product cards / A+ style content higgsfield-marketplace-cards Main image, secondary images, and A+ style modules with hidden marketplace prompt templates
Branded ad video / UGC / unboxing / TV spot higgsfield-generate Marketing Studio mode with avatars + products + optional hooks/settings
Analyze a video's hook / attention / virality potential higgsfield-generate Uses Virality Predictor (brain_activity) with --video; returns score metrics plus an Open report link
Train a custom face identity higgsfield-soul-id 5–20 photos, returns reference_id
Image-to-video animation higgsfield-generate Prefer seedance_2_5 with --mode omni_reference --start-image; kling3_0_turbo when the user asks for cheaper
Build / edit / deploy a website, web app, landing page, or dashboard higgsfield-websites Full-stack React 19 + TanStack Start on Cloudflare; higgsfield website create/repo-access/deploy
Create a narrated explainer from a topic or document higgsfield-faceless-video Generates scenes and narration, then assembles a finished video locally
Create a YouTube thumbnail or Shorts/Instagram video cover higgsfield-youtube-thumbnail Nano Banana Pro 4K main render, optional face/logo references, controlled variants, and Seedream tweaks
Build and deploy a playable browser game higgsfield-websites --type game: game design, assets, implementation, QA, deploy, and optional publish

License

MIT — see LICENSE.

skills (higgsfield-ai/skills)

正在搜索数据库…

这是什么?Higgsfield AI Skills 是一组基于 Markdown 的 AI 智能体技能,用于生成图片、视频、3D 和音频,并通过 Higgsfield AI 平台搭建完整的视觉品牌体系(包括标志、品牌手册、广告视频和网站)。
这是一个独立应用,还是我的编程智能体会加载的技能文件?它不是独立应用,而是一组 Markdown 技能文件,供 Claude Code、Cursor、Codex 等智能体加载和调用;每个技能都封装了对 Higgsfield CLI 的调用。
这些技能该怎么安装?任选一种安装方式:'npx skills'(推荐,支持多种智能体)、通过 GitHub CLI v2.90+ 使用 'gh skill install'、Claude Code 插件市场,或通用安装脚本;更多细节见 INSTALL.md,由智能体执行安装的指南见 INSTALLFORAGENTS.md。
使用这些技能需要 Higgsfield 账号吗?需要。无论采用哪种安装方式,都会为你设置好 Higgsfield CLI 及其身份验证,因为实际的生成任务运行在 Higgsfield 的后端模型上。
这些技能是各自独立运行,还是相互依赖?部分技能会串联使用:Soul ID 训练会把 reference id 输入 generate;brandkit 则把多个模型串成一个可感知审批流程的系统。而 product-photoshoot 和 marketplace-cards 是自包含的,因为后端会在提交任务前先优化提示词。
哪些编程智能体可以运行这些技能?它们面向 Claude Code、Cursor、Codex,以及其他能够加载基于 Markdown 技能的 AI 编程智能体;在 Claude Code 中可通过斜杠命令调用。
什么是 Marketing Studio?它是 higgsfield-generate 内的一套模式,用于制作品牌广告视频,提供 9 种模式,如 ugc、productshowcase、productreview 和 tvspot,可结合数字人形象、产品,以及可选的开场钩子和设置来制作。
什么是 Virality Predictor?这是 higgsfield-generate 内的一项评分功能(brainactivity),用于分析已完成视频的开场钩子、注意力和传播潜力,返回评分指标以及一个可直接打开的报告链接。
higgsfield-soul-id 有什么作用?它使用 5 到 20 张照片训练一个可反复使用、人脸还原度高的身份模型,并返回一个 referenceid,之后的生成任务可借此在不同输出中保持人脸一致。
这些技能能搭建并部署网站吗?可以。higgsfield-websites 可以创建、编辑并部署一个全栈 React 19 + TanStack Start 网站(甚至可以是可玩的浏览器游戏),整体以一个 Cloudflare Worker 的形式提供服务,并使用 D1、R2、KV 和 Durable Objects。
大小
0.9 MB
星标
2K
更新
提交Nov 05
时长
7个月
提交 · 90 天