Nanobanana Pro

Nano Banana Pro with auto model fallback — generate/edit images via Gemini Image API. Run via: uv run {baseDir}/scripts/generate_image.py --prompt 'desc' --filename 'out.png' [--resolution 1K|2K|4K] [-i input.png]. Supports text-to-image + image-to-image (up to 14); 1K/2K/4K. Fallback chain: gemini-2.5-flash-image → gemini-2.0-flash-exp. MUST use uv run, not python3.

Security Scan

The skill's code, instructions, and required environment (uv + GEMINI_API_KEY) are consistent with its stated purpose (image generation via the Gemini Image API with model fallback).

Purpose & Capability

Name/description (image generation/editing via Gemini) match the actual files and declared requirements. The script uses the Gemini client, accepts an API key, and requires the 'uv' runner as described. Requiring GEMINI_API_KEY and the uv binary is proportional to the stated functionality.

Instruction Scope

SKILL.md and the included script instruct the agent to run the bundled script with 'uv run', operate on user-supplied input images, and save an output file (printing a MEDIA: line). The instructions do not request unrelated files, extra credentials, or exfiltration endpoints. The script only reads input image paths provided by the user and writes the specified output file.

Install Mechanism

Install uses a Homebrew formula to install 'uv', a known runner. There are no downloads from arbitrary/personal URLs or extract steps that write unknown code to disk. 'uv' will manage Python dependencies at runtime (expected for this project).

Credentials

Only GEMINI_API_KEY is required (primaryEnv). An optional env var (NANOBANANA_FALLBACK_MODELS) is documented for configuration. No unrelated secrets or excessive environment access are requested.

Persistence & Privilege

The skill is not always-included and does not request elevated or cross-skill config changes. It does not modify other skills' settings or request persistent system-wide privileges.

Assessment

This skill appears coherent and implements an image-generation wrapper around the Gemini API. Before installing, confirm you trust the GEMINI API endpoint and the author: the script will send prompts and (optionally) user-supplied image bytes to Google’s Gemini service using the GEMINI_API_KEY. Keep your API key private (avoid passing it on shared CLI history), and consider setting limited-scope credentials if the provider supports them. Review the public repo (homepage) if you want further assurance. Installing 'uv' via Homebrew is expected; inspect what 'uv' will install on your platform. If you plan to supply local images, ensure you only pass files you intend to upload to the remote API.

Nano Banana Pro with Fallback

Use the bundled script to generate or edit images. Automatically falls back through multiple Gemini models if one fails.

⚠️ IMPORTANT: MUST use uv run or the generate wrapper. Do NOT use python3 directly — dependencies won't be available.

Generate (option A: wrapper script)

{baseDir}/scripts/generate --prompt "your image description" --filename "output.png" --resolution 1K

Generate (option B: uv run)

uv run {baseDir}/scripts/generate_image.py --prompt "your image description" --filename "output.png" --resolution 1K

Edit (single image)

uv run {baseDir}/scripts/generate_image.py --prompt "edit instructions" --filename "output.png" -i "/path/in.png" --resolution 2K

Multi-image composition (up to 14 images)

uv run {baseDir}/scripts/generate_image.py --prompt "combine these into one scene" --filename "output.png" -i img1.png -i img2.png -i img3.png

API key

• GEMINI_API_KEY env var
• Or set skills."nanobanana-pro-fallback".apiKey / skills."nanobanana-pro-fallback".env.GEMINI_API_KEY in ~/.openclaw/openclaw.json

Notes

• Resolutions: 1K (default), 2K, 4K.
• Models tried in order: gemini-2.5-flash-image → gemini-2.0-flash-exp-image-generation (configurable via NANOBANANA_FALLBACK_MODELS env var).
• Use timestamps in filenames: yyyy-mm-dd-hh-mm-ss-name.png.
• The script prints a MEDIA: line for OpenClaw to auto-attach on supported chat providers.
• Do not read the image back; report the saved path only.