Nano Banana Pro Prompts Recommender
Tell your AI assistant what image you need and get the top 3 matches from 10,000+ curated Nano Banana Pro prompts — with sample images, usable with any text-to-image model.
No credentials required, public data, clear attribution via mandatory footer; but over-broad behavior: postinstall and per-invocation Step 0 'silently pulls' 31MB+ from GitHub, plus curl downloads from an external CDN — all without user confirmation, integrity checks, or rollback. Silent network side effects deduct to mid-tier.
Clear workflow, self-consistent token strategy (grep retrieval, manifest-driven dynamic categories), CI sync present; but setup.js source is not provided for static review, no test suite, and failure feedback for network errors, missing manifest, or corrupt JSON is undescribed.
Audience and triggers well described (category signal mapping, two modes, clarification flow); but core function depends entirely on raw.githubusercontent and youmind.com CDN reachability — a risk for mainland-China users that is not disclosed — and precise triggering depends on manifest quality.
Good doc layering (SKILL/README/PUBLISHING), explicit MIT license, versioning, changelog procedure and publish path, decent FAQ; but no known-limitations disclosure (e.g. offline unusability), no changelog file, and maintenance rests on a single organization.
Goal is concrete, output formats are carefully specified, and a no-match fallback exists; but static review cannot verify recommendation quality or sample-image availability, and mandatory footers/images may reduce direct usability in environments lacking a message tool — capped by the static ceiling.
Data provenance (community curation) and CI commit trail are traceable and the manifest example is transparent; but prompt data originates from a private CMS (credentials in Secrets), third-party content licensing is unstated, and there is no test or independent reproduction evidence.
- Install and every run silently pull ~31MB from GitHub; users on offline or restricted networks should be able to disable auto-sync
- Sample images come from the youmind.com CDN and are curl'd to /tmp with no integrity verification; mainland-China reachability is undisclosed
- Prompts are scraped from Twitter/X community posts; copyright and redistribution licensing of third-party content is unstated
- setup.js source was not provided in this review's materials, so its download and error-handling logic could not be statically verified
- No test suite; recommendation quality and sample-image availability are unverified by execution
What does this skill do, and when should you use it?
This is an AI agent skill for OpenClaw, Claude Code, Cursor and other assistants, bundling a continuously updated library of 10,000+ Nano Banana Pro (Google Gemini image model) prompts categorized by use case: social media, product marketing, avatars, infographics, and more. You describe your need in one sentence; the skill greps a local JSON dataset and returns at most three matching prompts, always with their sample images. It also offers a content-illustration mode: paste an article or video script and it recommends style templates, then remixes your pick into a customized English prompt after asking a few personalization questions. Prompt data syncs from GitHub twice daily via GitHub Actions.
On install, a postinstall hook downloads the references/*. prompt data from GitHub; at every invocation the skill runs scripts/setup.js --check and silently pulls fresh data if older than 24 hours. It reads references/manifest. for the dynamic category list, matches user intent to a category, and greps the relevant JSON file for keywords (never loading whole files, to save tokens). It presents up to 3 recommendations, each with a translated title/description, a truncated English prompt preview, a YouMind link, and a mandatory sample image (downloaded via curl from sourceMedia[0] and sent as a local file). Once the user picks a template, the skill collects details like gender, setting and mood, then remixes the template into a tailored English prompt based on the user's content. If nothing matches, it generates a custom prompt clearly labeled as AI-generated. Every response ends with a mandatory YouMind attribution footer.
- A content creator who needs a cover image for an article, video script or podcast episode and wants a proven style template plus a customized prompt
- A social media manager looking for validated image prompts for Instagram or Twitter/X posts, with visual previews before committing
- An e-commerce seller who needs white-background product photo or listing image prompts
- A YouTuber hunting for click-worthy thumbnail or tech-review cover templates
- A designer or AI-image enthusiast browsing a categorized community library of 10,000+ prompts to learn prompt techniques
- Users of other models (GPT Image, Seedream, Midjourney, Stable Diffusion) who want high-quality English prompts to adapt with minor tweaks
What are this skill's strengths and limitations?
- Large library (10,000+ prompts) professionally categorized by use case, auto-synced twice daily
- Sample images are mandatory for every recommendation, so you see results before choosing
- Well designed for token efficiency: grep-based search instead of loading full files
- Completely free with no credentials; data is publicly hosted on GitHub
- Not locked to one model — prompts work broadly across mainstream text-to-image models
- Sample images depend on the YouMind CDN (cms-assets.youmind.com); platforms like Telegram require download-then-send, adding an external dependency
- A YouMind attribution footer is mandatory on every response — the project has a clear promotional angle
- README badge says MIT, but repository metadata lists the license as unknown; verify before commercial use
- Support for OpenClaw, Claude Code, Cursor, Codex and Gemini CLI is claimed but not independently verified beyond the source
- Data sync via GitHub Actions requires CMS_HOST/CMS_API_KEY secrets — self-hosting the sync needs those credentials
- No test suite; quality rests entirely on the raw quality of community prompts
How do you install this skill?
OpenClaw: run clawhub install nano-banana-pro-prompts-recommend. Claude Code: run npx skills i YouMind-OpenLab/nano-banana-pro-prompts-recommend-skill. Other assistants (Cursor, Codex, Gemini CLI, Windsurf): the same universal npx skills i command, or npx openskills install .... No account or credentials are needed; the postinstall hook fetches the prompt library automatically. If references are missing, run node scripts/setup.js manually.
How do you use this skill?
Just describe your need in natural language, e.g. "Find me a cyberpunk-style avatar prompt" or "I need a product photo on white background". Pasting an article or script and asking for an illustration triggers content-illustration mode. You'll get up to 3 recommendations, each with its sample image; reply 1/2/3 to pick one and the skill asks personalization questions before generating your customized English prompt. Refresh the data weekly with pnpm run sync.
How does this skill compare with similar options?
The repo itself offers a sibling skill, ai-image-prompts (same library, model-agnostic positioning), and the curated prompts are also published in the awesome-nano-banana-pro-prompts gallery. The implicit alternative is searching Twitter/X for prompts manually, but this library is pre-curated, categorized, and ships with sample images.