Nano Banana — Image Generation

Image generation script supporting three models. Requires sharp for input image compression (auto-installed on first run).

Models

Generate an image

node {baseDir}/scripts/generate-image.js --prompt "a cat sitting on mars" --filename "cat-on-mars.png"

Edit a single image

node {baseDir}/scripts/generate-image.js \
  --prompt "make the sky purple" \
  --filename "edited.png" \
  -i "/path/to/input.png" \
  --model nano-banana-pro

Multi-image composition (up to 14 images)

node {baseDir}/scripts/generate-image.js \
  --prompt "combine these into a collage" \
  --filename "collage.png" \
  -i img1.png -i img2.png -i img3.png

Options

API key

The API key is pre-configured on this machine. No flags or environment variables needed.

Input image handling

All input images are sent as inline base64. Images over 500 KB are automatically compressed to JPEG and resized to fit under the limit. This keeps requests fast and avoids File API auth issues with the enterprise endpoint.

Output

Relative filenames are saved to $OPENCLAW_STATE_DIR/media/outbound/{slugid}/nano-banana/{filename}. Absolute paths are used as-is. Absolute paths are used as-is. Use timestamps in filenames to avoid overwrites: cat-on-mars-20260304-165000.png.

Sending images to the user

The script prints a MEDIA: <absolute-path> line on stdout. You MUST include this exact MEDIA: line in your reply text so the image is delivered as an attachment in Discord/Slack/chat.

Example reply:

Here's your image!
MEDIA: /Users/alche/.openclaw/media/outbound/my-bot/nano-banana/cat-on-mars.png

Rules:

• Copy the MEDIA: line from the script output into your reply verbatim — this is how images get sent
• Do NOT read the generated image back with the read tool
• Do NOT try to base64 encode or manually attach the image
• The MEDIA: line must be on its own line in your response