IMINI ImageGen

Generate images through the imini image generation API using a local API key from environment variables. This skill supports text-to-image and image-guided generation with public image URLs, asks the user to choose a model when they did not specify one, then polls the async task until final images are ready and saved into the workspace.

When To Use

Use this skill when:

• The user wants to generate an image, poster, cover, concept image, or other raster output.
• The user provides a prompt and wants you to actually call the imini API.
• The user provides one or more public image URLs as references for style or content guidance.

Do not use this skill when:

• The user wants SVG, icons, diagrams, or HTML/CSS visuals.
• The task is only to brainstorm or rewrite prompts without actually generating images.
• The only reference input is a local file path; v1 supports public URLs only.

Configuration

The script reads configuration from environment variables. In OpenClaw, declare IMINI_IMAGE_API_KEY in the skill config so the registry and runtime requirements stay aligned.

• IMINI_IMAGE_API_KEY (required)
• IMINI_IMAGE_API_BASE_URL (optional, default https://openapi.imini.ai/imini/router)
• IMINI_IMAGE_TIMEOUT (optional, default 300)
• IMINI_IMAGE_POLL_INTERVAL (optional, default 2)

Do not put real keys into this skill directory, SKILL.md, or command arguments. Setup details live in references/config.md.

Workflow

1. Confirm the request is actual image generation, not vector/UI drawing.
2. Collect the prompt.
3. Optionally collect:
   • model
   • aspect_ratio
   • public image URLs
   • reference mode if the user clearly wants style instead of default asset
4. If the user did not specify a model, ask before generating:

Which model do you want to use?
1. google/nano-banana: faster and lower cost for standard image generation
2. google/nano-banana-pro: higher quality and higher resolution support
3. google/nano-banana-2: most capable option with more aspect ratios and resolutions

If you are not sure, I recommend starting with google/nano-banana.

Do not call the API until the user chooses a model, unless they explicitly ask you to use the default recommendation.

5. Run the bundled script:

python3 /absolute/path/to/imini-imagegen/scripts/generate_image.py \
  --model google/nano-banana \
  --prompt "A shiba inu wearing a spacesuit and standing on the moon, 3D cartoon style" \
  --aspect-ratio 1:1 \
  --out-dir output/imagegen

With references:

python3 /absolute/path/to/imini-imagegen/scripts/generate_image.py \
  --model google/nano-banana \
  --prompt "Keep the main composition and turn it into a cinematic poster with dramatic lighting" \
  --image-url "https://example.com/ref1.jpg" \
  --reference-type asset \
  --out-dir output/imagegen

6. Wait for the script to:
   • submit POST /v1/images/generate
   • poll GET /v1/images/tasks/{task_id}
   • download final images[].url results
7. Report:
   • final saved file paths
   • task status
   • failure reason if the task fails

Input Rules

• prompt is required.
• Supported models:
  • google/nano-banana
  • google/nano-banana-pro
  • google/nano-banana-2
• If the user did not specify a model, ask them to choose before generating.
• If the user asks for a recommendation, prefer google/nano-banana as the default low-friction option.
• resolution defaults to 1K. Higher values should only be used when the chosen model supports them.
• aspect_ratio is optional. Allowed values are documented in references/api.md.
• --image-url count limits depend on the chosen model.
• v1 accepts only public image URLs for references.
• reference_type defaults to asset; use style only when the user clearly wants style transfer behavior.

Failure Handling

If generation fails, check in this order:

1. IMINI_IMAGE_API_KEY is present.
2. The API returned auth or validation errors.
3. Reference URLs are public and reachable.
4. The task status changed to failed.
5. The task timed out before completion.
6. The task succeeded but images was empty or downloads failed.

For API shapes, status values, and exact request fields, read references/api.md.