| name | gemini-image-generator |
| description | Generate, edit, or transform images with Gemini Nano Banana using bundled Python scripts (Flash or Pro) including aspect ratio, resolution, image-to-image edits, logo overlays, and reference images. Use when users request image generation, image edits, image-to-image transformations, logo placement, or specific aspect ratios or resolutions. |
Gemini Image Generator
Use this skill to turn a user prompt (and optional images) into Gemini image generation calls via the bundled Python scripts.
Workflow
- Collect the user prompt and any images (local paths or URLs).
- Infer the operation mode and translate parameters into CLI flags.
- Run the appropriate script to generate or edit images.
- Return the output file paths.
Defaults and Rules
- Default model:
gemini-2.5-flash-image(CLI valueflash). - Default aspect ratio:
9:16. - Default count:
1(max3). - Default image size:
1K, but only apply it for the Pro model. - If the user specifies a size (
1K|2K|4K), switch to Pro (gemini-3-pro-image-preview). - If the user explicitly asks for Pro or higher quality, use Pro.
- If the user supplies multiple reference images (2+), switch to Pro.
- Logo overlay always uses Pro (even if the user asks for Flash).
- Only set
--sizewhen using Pro.
Allowed Values
- Aspect ratios:
1:1,2:3,3:2,3:4,4:3,4:5,5:4,9:16,16:9,21:9. - Image sizes (Pro only):
1K,2K,4K. - Reference images (Pro): up to 14 total. Gemini guidance: up to 6 object images + up to 5 human images.
Script
Scripts
Text-to-image
Run:
python scripts/generate_image.py \
--prompt "<user prompt>" \
--aspect 9:16 \
--count 1 \
--model flash \
--out-dir outputs
Only add flags when the user asks for them. The script reads GEMINI_API_KEY from the environment.
Image editing / image-to-image
Use when the user supplies a base image to edit or transform.
Run:
python scripts/edit_image.py \
--input /path/to/base.png \
--prompt "<edit instructions>" \
--reference /path/to/ref1.png \
--reference https://example.com/ref2.png \
--aspect 9:16 \
--count 1 \
--model flash \
--out-dir outputs
Logo overlay
Use when the user wants to place a logo onto a base image.
Run:
python scripts/logo_overlay.py \
--base /path/to/base.png \
--logo /path/to/logo.png \
--aspect 9:16 \
--count 1 \
--model pro \
--out-dir outputs
Examples
User: "Generate a portrait of a dancer in a foggy forest." Claude:
- Use defaults (flash, 9:16, count 1).
- Run:
python scripts/generate_image.py --prompt "Generate a portrait of a dancer in a foggy forest."
User: "Make a 2K 16:9 cinematic still of a neon city, give me 3 options." Claude:
- Use Pro with size 2K, aspect 16:9, count 3.
- Run:
python scripts/generate_image.py --prompt "Make a 2K 16:9 cinematic still of a neon city" --aspect 16:9 --size 2K --count 3 --model pro
User: "Edit this image to remove the background and make it studio white." (with one image) Claude:
- Use edit script with Flash.
- Run:
python scripts/edit_image.py --input /path/to/image.png --prompt "Remove the background and make it studio white."
User: "Put this logo on the shirt in the photo." (with base + logo images) Claude:
- Use logo overlay script (Pro).
- Run:
python scripts/logo_overlay.py --base /path/to/photo.png --logo /path/to/logo.png
Notes
- If the script fails with a missing module, install
google-genaiand retry. - Dependencies live in
scripts/requirements.txt(install withpip install -r scripts/requirements.txt). - Output files are written into the
outputs/directory using timestamped names. - For prompt best practices and templates, read
references/prompt-guide.md. - For logo-specific guidance, read
references/logo-overlay.md. - For edit/image-to-image guidance, read
references/image-editing.md. - For watermarking guidance, read
references/watermarking.md.