tuoxie2046/claude-code-research-skills/nano-banana
- community
Checking that the SKILL.md file is still in the repository.
Generate images from the terminal with Google's nano-banana (Gemini image model) via the Antigravity CLI (`agy`), billed to the signed-in Google account. Use whenever the user asks to create / draw / generate an image, picture, icon, logo, illustration, poster, infographic, or diagram specifically with nano-banana / Gemini / Antigravity — or wants rich, text-heavy infographics. For the OpenAI gpt-image-2 path use the separate `gpt-image` skill instead.
Install
npx vskill@latest install tuoxie2046/claude-code-research-skills/nano-bananaSKILL.md
nano-banana — image generation via Antigravity (agy)
Wraps the agy (Google Antigravity) agent so Claude Code can generate images with nano-banana. Unlike a plain CLI, image generation here is an agent tool (generate_image) invoked through natural language — the wrapper builds a strict instruction, runs agy -p, then verifies the output file actually landed (with recovery from agy's brain/ working dir). Auth is the Google account signed into agy.
All commands go through the helper:
Workflow
- Preflight once per session:
nano-banana.sh check. If it reports no creds /
auth, the user must run agy once interactively to log in (browser OAuth) — you cannot do this for them. Token expiry surfaces at generate time as an auth error.
- Generate with an explicit
--outputPNG path. The helper prints
IMAGE\t<path>\t<WxH> on success.
- Always `Read` the produced PNG to view it and verify, then report the path.
- Iterate by re-running with a refined prompt.
Commands
Generate (text → PNG):
Batch (many prompts → many images; one per prompt, continue-on-error, manifest):
Name an output by prefixing a line with stem<TAB>prompt (else 001.png, …). Default --outdir is ./nb-batch-<timestamp>/; read <outdir>/manifest.tsv for the stem→status→path→prompt map. Per-image failures are recorded and skipped, but an auth failure aborts the whole batch (no point retrying every prompt against dead auth) — re-login with agy and re-run.
How it works / behaviors
- Underlying tool is
generate_image(nano-banana is the model). There is **no
agy image subcommand** — the agent calls the tool from the prompt.
- The agent generates a JPEG into `~/.gemini/antigravity-cli/brain/<conv-id>/`, then
converts to PNG at your --output. The wrapper forces an absolute path and, if the agent doesn't save there, recovers the newest brain image and converts it via sips.
- Output paths are absolutized (agy resets cwd). Parent dirs are created.
- One retry on a no-file result; auth/eligibility errors fail fast with guidance.
- Non-destructive: success requires the output file to actually change this run
(checked via its inode:mtime:size), so a pre-existing/stale file is never reported as a fresh result and is not deleted if generation fails.
⚠️ Safety
This path drives an autonomous agent (agy -p … --dangerously-skip-permissions): tool calls and file writes are auto-approved. A prompt is therefore not just image text — a malicious or careless prompt could instruct the agent to run other tools or touch other files. Only pass trusted prompt text (and trusted --prompts-file contents). For the non-agent, capability-scoped path, prefer the gpt-image skill.
Strengths & limitations (vs gpt-image-2)
- Strengths: excellent in-image text rendering; tends to produce rich, fully
fleshed-out infographics (adds detail, gradients, sub-labels). Great for posters and content-heavy diagrams; strong at consistent image editing (via prompt + a file path).
- Limitations:
- No precise size control through the agent — it returns its own dimensions
(often ~16:9). If you need an exact size, post-process with sips -z H W.
- Slower than gpt-image-2 (it's an agent turn: tool call + format convert).
- Tends to elaborate — if you want strictly minimal output, say "minimal, no
extra text, exactly three boxes" etc.
- Needs a live
agylogin; tokens expire and require interactive re-auth.
Troubleshooting
agy ... not found→ install Antigravity (https://antigravity.google) or ensure
~/.local/bin is on PATH.
- auth /
IneligibleTier/ "Authentication required" → runagyonce interactively to
log in; the standalone gemini CLI is not usable on free-tier accounts.
- No file produced → check stderr for the agy transcript; re-run (one retry is automatic).
Security
- Security scanPassed, 95 of 100Pattern checks on the skill's files.
- Model reviewPassed
Quality evaluation
- Task evaluationNot runNo evaluation yet. Run one in Skill Studio.
Source
- SKILL.md in the repositoryCheckingskills/nano-banana/SKILL.md
Popularity
- Repository stars
- 5
- Repository forks
- 1
- 7-day trend score
- 9
Stars and forks are for the whole GitHub repository, not this skill alone. Not refreshed yet.
Works with
Any agent vskill installs to: 56 targets in all. These 39 have a page on this site.
- Claude Code
- Cursor
- GitHub Copilot
- Windsurf
- Codex
- Gemini CLI
- Cline
- Amp
- Roo Code
- Goose
- Continue
- Kilo Code
- Junie
- Trae
- OpenHands
- Qwen Code
- AI Agent
- Tabnine Chat
- Sourcegraph Cody
- Amazon Q
- Firebase Genkit
- Sweep
- Devin
- Phind
- Mentat
- Claude Desktop
- Aider
- GPT Pilot
- Plandex
- bolt.new
- v0
- Replit Agent
- Anse
- TypingMind
- BetterChatGPT
- LibreChat
- Warp AI
- Pieces
- Fig AI
Embed badge
[](https://verified-skill.com/skills/tuoxie2046/claude-code-research-skills/nano-banana)