ai-media-generator skill
A skill that writes prompts for video, image and music generation on 14 platforms and can submit them in the browser
- 266 ★ on the repository
- Last commit
- MIT
Broad access
Why this level
- Through a browser tool the skill can press buttons on sites where you are already signed in
- Generation on paid platforms spends credits, although the skill stops before payment and publishing
- Downloaded results and images you upload pass through third-party generation services
Install
- Works in
In your terminal, with SkillFoxx CLI
npx skillfoxx add skills/ai-media-generatorDetects the agents on your machine, shows the access level and pins the version.
Other ways to install
Access and checks
How we reviewBroad accessThe tool can change data, run commands or handle secrets.
- Through a browser tool the skill can press buttons on sites where you are already signed in
- Generation on paid platforms spends credits, although the skill stops before payment and publishing
- Downloaded results and images you upload pass through third-party generation services
This is third-party code. Review the repository files before installing.
What it does
ai-media-generator helps an agent get quality results from generative models from a single sentence. You write a task like "a short film about an ancient general charging on horseback", and the skill picks a platform, writes a prompt for it and, if needed, submits it to the site itself. The references cover Midjourney, Flux, Veo 3.1, Kling 3.0, Seedance 2.0 and 2.5, Runway Gen-4.5, Suno v5.5, Ideogram, Seedream, Nano Banana Pro, Vidu, Stable Diffusion, OiiOii and others. Sora 2 is marked as discontinued. The core idea: every model has its own favourite words. Midjourney likes comma-separated details, Flux values a paragraph of 80 to 200 words, Seedance understands concrete imagery in Chinese plus a constraints tail better, and Veo is the only one that really voices sound tags. So by a hard rule each prompt carries 5 to 8 terms from vocabularies of director, cinematographer, lens, film stock, lighting and color grading. Empty words such as `beautiful` and `masterpiece` are banned. Autopilot mode splits a request into nine slots (medium, length, aspect ratio, subject, style, character, scene, audio, language), fills sensible defaults, picks a platform from a matrix and writes a logline and storyboard. A short window for edits comes before execution. When the skill drives a site through a browser MCP, it stops before a paid plan, submission, download and publishing, and before sensitive content. The skill holds more than 55 files: platform and camera references, storyboard and music video templates, a negative prompt bank and site profiles for OiiOii, Flow, Kling and Suno.
Who it is for. For creators of videos, clips and illustrations who work with several generative models and do not want to study each one's prompt quirks.
Good fit when
- You need a clip on Seedance, Veo or Kling and the prompt must follow that model's quirks
- It is unclear which platform fits the task: the `selector.md` table helps choose between Veo, Kling, Runway and Seedance
- You need a storyboard or a music video: the set includes `storyboard.md` and `music-video.md` templates
- You want the agent to submit a prompt through the browser to Flow, Kling or Suno, stopping before payment
- You need prompts for advertising, real estate, food or cars: Seedance has ready domain playbooks
Not a fit when
- You have no accounts on the target platforms: the skill writes a prompt but cannot sign in or pay for you
- You need an AI series or short drama with episodes: the author has a separate skill, `ai-short-drama`, for that
- Agent context is limited: the full set of 55 or more files is heavy, so enable the token-saving mode
- You need generation on your own computer: the skill works with sites and services and does not run local models
Example request
Make a cinematic 15 second clip: an ancient general on a black horse charges at dawn. Pick the platform and write the prompt, and show it to me before submitting.Limitations
- The skill writes prompts and drives sites through a browser tool or MCP connected to the agent. Without it only text drafting remains.
- The platforms need your accounts signed in within the same browser. The skill does not buy paid plans, balance or subscriptions, and it stops at a payment wall.
- Site profiles are verified for OiiOii, Google Flow with Veo 3.1, Kling 3.0 and Suno v5.5. Midjourney, Seedream, Runway, Sora, Vidu and Ideogram have only stubs.
- Site interfaces change, so the steps in the profiles age over time.
- References and the description are mostly in Traditional Chinese with English inserts. Sora 2 is discontinued: the app and site closed on 26 April 2026 and the API ran until 24 September.
- The skill's reference describes Suno v5.5; since 9 September 2026 Suno's current version is v6, so the skill may not know the new parameters.
How to disable. Delete the `ai-media-generator` folder from `~/.claude/skills/`, `~/.codex/skills/` or the project's `.claude/skills/`. The skill leaves no data or background processes.
FAQ
How do I install ai-media-generator in Claude Code?
Clone the repository with `git clone https://github.com/Hao0321/ai-media-generator.git ~/.claude/skills/ai-media-generator` and restart Claude Code. For Codex use the path `~/.codex/skills/ai-media-generator`.
Which platforms does the skill support?
References exist for Midjourney, Flux, Veo, Kling, Seedance, Runway, Suno, Ideogram, Seedream, Nano Banana Pro, Vidu, Stable Diffusion and OiiOii. Browser automation is verified on OiiOii, Flow, Kling and Suno.
Why do the prompts avoid words like masterpiece and 8K?
The skill bans empty boosters and requires 5 to 8 precise terms: director and cinematographer names, lens, film stock, lighting, grading. For newer models such as Seedance 2.0, `cinematic` and 4K count as working tokens, and platform choice accounts for that.
Can the skill spend my money?
Autopilot stops before a paid plan, final submission, download and publishing. With no reply and a paid action it waits, while a free action starts after thirty seconds.
What about the skill's large size?
The skill has a token-saving mode: it reads only needed reference files through a quick lookup table and does not load all 55 or more files at once.
Discussion
No questions or reviews yet.
Related
SkillsTwo skills for generative video and images: video writes Seedance, Kling and Veo prompts by dramaturgy rules, image picks the model and its syntax
A skill that turns an idea, a link or a screen recording into finished video: HeyGen avatar, b-roll, captions and quality checks
A skill that searches a library of 10,000+ image prompts, shows sample images and tailors the chosen prompt to your article or script
A skill for Cursor, Claude Code and Codex: from a product name or URL it builds a 4K promo with music, voice-over and checks
A Claude Code skill that turns an idea into a five-stage prompt for Sora, Kling, Veo and Seedance, with templates for 21 genres
A skill that turns a topic into a narrated 4K video: research, script, local TTS, Remotion composition and an FFmpeg mix
Comparison with similar tools
Entries that share the main search query, and how they differ.
| Entry | Official | Repository stars | Main difference |
|---|---|---|---|
| ai-media-generator (this entry) | no | 266 | |
| Seedance 2.0 Prompt Writing Skills | no | 4,376 | teaches prompts for Seedance 2.0 only, while here 14 platforms and an autopilot are covered |
| Prompt Master | no | 15,225 | writes prompts for tools in general, while here the focus is video, images and music with site profiles |
README in short
The page in Chinese and English explains why prompts are not portable between models, shows an example and the file layout. It continues with key rules, autopilot mode, install, requirements and a table of verified platforms.
README badge
Original SKILL.md
--- name: ai-media-generator description: 為使用者產生高品質的 AI 生圖、生影片、生音樂提示詞，並在需要時透過瀏覽器自動化實際送到目標平台。涵蓋 OiiOii、Kling 3.0/O-series、Seedance 2.0/2.5、Suno v5.5、Seedream 5.0/4.0、Vidu Q3、Midjourney V8.1、Flux 1.1 Pro / Kontext、Runway Gen-4.5 / Aleph、Google Veo 3.1、Ideogram 3、Nano Banana Pro、Stable Diffusion 3.5（⚠️ OpenAI Sora 2 已於 2026-04-26 停運，API 撐到 2026-09-24，預設改推 Runway/Veo/Kling）。只要使用者提到「AI 生圖」「AI 影片」「AI 音樂」「做 MV」「做 storyboard」「寫 prompt 給 XXX」「我想用 Kling/Suno/Midjourney/Runway/Veo...」「幫我操作 OiiOii / 即夢 / 可靈」「txt2img / img2video / 文生圖 / 文生影片 / 圖生影片」「角色一致性」「多鏡頭分鏡」「運鏡」「結果有瑕疵 / 不夠精緻 / 怎麼修」，或任何跟上述平台或影像/影片/音樂生成工作流相關的任務，都要用這個 skill。即使他們沒講明平台，只要任務是要餵給某個生成模型的 prompt，就用這個 skill 幫他們選對的平台、寫對的格式。 --- # AI Media Generator > **Codex compatibility:** References that still say “Claude in Chrome” or “Chrome MCP” describe legacy browser behavior. On Codex, load `chrome:control-chrome` and use the currently available Chrome tools; never invent or call legacy `mcp__Claude_in_Chrome__*`, `javascript_tool`, or `browser_batch` names. 幫使用者把想法變成高品質的 AI 生成內容 (圖片、影片、音樂)，核心工作是 **寫對每個平台的 prompt** 以及 **必要時自動操作網站**。 **AI 短劇邊界：** 使用者要做 AI 短劇／漫劇、重生／系統／逆襲／隱藏大佬、分集連載或整季架構時，先由 [`ai-short-drama`](https://github.com/Hao0321/ai-short-drama) 建題材、Bible、pilot、分集狀態與 production pack；本 Skill 只接下游角色圖、場景圖、逐鏡影片／聲音 prompt 與模型執行，不自行重寫季弧。 ## 🤖 Auto-Pilot Mode (超傻瓜一句話到成品) **觸發：** 使用者說「幫我做 X」「做一個 X」「生個 X」「我要 X」這類命令式、且 X 含媒體類型 (圖/影片/音樂/MV/短片/海報/動畫) 或風格關鍵字 → **直接進 auto-pilot 不問**。 **流程 (見 [templates/auto-pilot.md](templates/auto-pilot.md) 完整版)：** 1. **Intent Parser** — 一句話拆 9 slot (媒體/長度/畫面比/主題/風格/角色/場景/音訊/語言) 2. **Fill Defaults** — 沒講的用預設 (video 預設 10s 16:9；動漫預設 Shinkai+Ghibli；電影預設 Deakins DP…) 3. **Platform Decider** — 按 [references/selector.md](references/selector.md) + quick pick 矩陣自動選 4. **Script + Storyboard Auto-gen** — 自動寫 logline + 角色卡 + 分鏡 5. **Prompt Crafting** — 強制套語彙庫 (見下方硬規則) 6. **Preview + Go** — 30-秒 override 視窗，使用者回「go/確認/OK」才執行；無回 + 需花錢 → 停；無回 + 免費 → 30 秒後自動繼續 7. **Execution + Report** — 按 [automation/click-protocol.md](automation/click-protocol.md) + [site-profiles/](automation/site-profiles/) 協議操作 **Auto-Pilot 必停的 checkpoint (不代做)：** - ⛔ 付費 paywall / 升級提示 - ⛔ 送出前最終一次確認 - ⛔ 下載 / 分享 / 公開發布 - ⛔ 敏感內容 (真人照片疑似裸露/暴力/政治人物) - ⛔ click-protocol.md 定義的「irreversible action」 **Auto-Pilot 禁止行為：** - ❌ 幫 Claude 已知的事問使用者 (「你要幾秒的影片？」— 10s 預設就好) - ❌ 一次拋多選項 — 選 1 個最佳 commit，讓使用者在 Preview 階段才決定改不改 - ❌ 寫純 generic 填詞 prompt (`beautiful / masterpiece / detailed`) — 違反下方硬規則（注意：`cinematic / 4K / 8K` 在新一代模型如 Seedance 2.0 是有效 token，看平台分流） - ❌ 代付款 **使用者自然語言 flags 可 override 預設：** 使用者一句話可帶「用 Kling / 5 秒 / 豎屏 / 免費 / Ghibli 風 / 有對白」等自然語言 flag，auto-pilot 掃 [templates/user-flags.md](templates/user-flags.md) 對照表自動套用。使用者不懂術語也 OK — 「做個抖音」「可愛一點」「夢幻」都有對應翻譯。 **⚡ Token-Efficient Mode (大專案必讀)：** 本 skill 55+ 檔 / 14000+ 行，全量讀會爆 context。Auto-Pilot 預設套 [templates/token-efficient-mode.md](templates/token-efficient-mode.md) 的 7 層策略 (lazy load / grep / 子代理 / preset 跳過 / cache polling)。**一般任務目標 ~25-40k tokens，不是 100k+**。豪華模式 (全量讀) 只在 benchmark / 學 skill / 陌生平台探索時啟用。 ## 🎯 快查表：3 秒找對檔案（更直覺入口）