v1.2.0 — nano-banana image generation skill

nano-banana — new skill

Wraps Google's Gemini image-generation API (Nano Banana family) so Claude / Claude Code can generate or edit images programmatically from any session. Before this, the only path was pasting prompts into the Nano Banana UI by hand.

Highlights

• Stdlib-only Python CLI — no requests, no google-genai dependency. Single-file generate.py.
• Default model: gemini-3.1-flash-image-preview (newest flash-tier image model, 16:9 native).
• Pro tier: --model nano-banana-pro-preview for higher fidelity hero/deck images.
• Stable fallback: --model gemini-2.5-flash-image if the preview goes flaky.
• Image edit / image-to-image: pass --input ref.png (repeatable for multi-reference composition).
• Documented model catalog in SKILL.md, including the gemini-3.5-flash text-only gotcha (returns finishReason: NO_IMAGE).

Setup

export GEMINI_API_KEY="..."   # https://aistudio.google.com/apikey
python3 ~/.claude/skills/nano-banana/generate.py \
  --prompt "editorial photo of …" \
  --out output.png

Tracking

• Issue: #11
• PR: #12

Full diff: v1.1.0...v1.2.0