Files
pi-skill/skills/nano-banana/SKILL.md
2026-05-25 16:41:08 +07:00

58 lines
2.2 KiB
Markdown
Raw Blame History

This file contains ambiguous Unicode characters
This file contains Unicode characters that might be confused with other characters. If you think that this is intentional, you can safely ignore this warning. Use the Escape button to reveal them.
---
name: nano-banana
description: Generate or edit images with OpenRouter using the google/gemini-3.1-flash-image-preview model. Use when the user asks for image generation, wants a new visual from a text prompt, or wants modifications to an existing screenshot or image and expects an image file back.
allowed-tools: Bash(node agent/skills/nano-banana/generate-image.js:*)
---
# Nano Banana Image Generation
Use this skill when the user wants an image back, either from a fresh prompt or from editing an existing image.
## Model
- Provider: OpenRouter
- Model: `google/gemini-3.1-flash-image-preview`
- Friendly name: Nano Banana
## Supported workflows
1. **Text-to-image**: user provides a prompt, you generate an image.
2. **Image edit**: user provides an image or screenshot plus instructions, you generate a modified image.
## Before you run
- Make sure the prompt is explicit enough to produce the desired image.
- If editing an image, first save or locate the input image path in the workspace.
- The script writes generated files to `.context/generated-images/` by default.
## Usage
### Generate a new image
```bash
node agent/skills/nano-banana/generate-image.js \
--prompt "A clean product hero shot of a matte black coffee grinder on a soft beige background, studio lighting" \
--output .context/generated-images/coffee-grinder.png
```
### Edit an existing image
```bash
node agent/skills/nano-banana/generate-image.js \
--prompt "Keep the layout the same, but change the CTA button to green, replace the headline with 'Ship faster', and make the page look more premium" \
--input-image path/to/screenshot.png \
--output .context/generated-images/edited-screenshot.png
```
## Response handling
- The script prints JSON with an `output` field containing the saved image path.
- After running it, tell the user the output file path.
- If the model returns no image, inspect the JSON warning and summarize the failure clearly.
## Notes
- Prefer PNG outputs unless the user asks for another format.
- Preserve existing composition when the user asks for modifications to a screenshot.
- If the users request is ambiguous, clarify what should change and what must remain untouched.