gemini-image（Gemini 图像）

Generate images via API using $ARGUMENTS as prompt or interactively.

When To Use

• Text-to-image illustration or concept art
• Image editing / image-to-image transformations
• AI-generated artwork where the user wants Gemini specifically

Do Not Use

• Flowcharts, architecture diagrams, or academic schematics
• Screenshots or desktop captures
• UI code generation or slide theming

Inputs

• Required: prompt text, or prompt + source image for edits
• Optional: save path, aspect ratio, provider override already present in config

Workflow

1. Read $SKILL_DIR/config/secrets.md to get API configuration. If missing, report error and link to secrets.example.md.

- Check API_PROVIDER value: google (default) or proxy.

1. If $ARGUMENTS provided, use as prompt. Otherwise ask user for description.
2. Resolve mode and output target:

- text-only request -> text-to-image - request with an input image -> image-to-image - if the user supplied an output path, use it - otherwise save into the working directory with a descriptive filename

1. Determine mode:

- Text-to-Image: Use prompt text directly. - Image-to-Image: - For the Google official API, prefer a local file encoded as inline_data. - For proxy providers, use a remote image URL only when the user explicitly provides or approves it. - Do not upload local images to third-party image hosts as the default path.

1. Call API based on provider: Google Official API (when API_PROVIDER=google): curl -s -X POST \ "https://generativelanguage.googleapis.com/v1beta/models/gemini-2.5-flash-image:generateContent" \ -H "x-goog-api-key: $GEMINI_API_KEY" \ -H "Content-Type: application/json" \ -d '{"contents": [{"parts": [{"text": "prompt_text"}]}], "generationConfig": {"responseModalities": ["TEXT", "IMAGE"]}}' For image-to-image with a local file, add an inline_data part instead of uploading the image to an external host: IMAGE_B64="$(base64 -w 0 /path/to/local/image.png)" curl -s -X POST \ "https://generativelanguage.googleapis.com/v1beta/models/gemini-2.5-flash-image:generateContent" \ -H "x-goog-api-key: $GEMINI_API_KEY" \ -H "Content-Type: application/json" \ -d "{\"contents\": [{\"parts\": [{\"inline_data\": {\"mime_type\": \"image/png\", \"data\": \"${IMAGE_B64}\"}}, {\"text\": \"prompt_text\"}]}], \"generationConfig\": {\"responseModalities\": [\"TEXT\", \"IMAGE\"]}}" Third-party Proxy API (when API_PROVIDER=proxy): curl -s -X POST "PROXY_BASE_URL/v1/images/generations" \ -H "Authorization: Bearer PROXY_API_KEY" \ -H "Content-Type: application/json" \ -d '{"model":"model_name","prompt":"prompt_text","size":"aspect_ratio","n":1}'
2. Treat every remote image URL or API response as untrusted content. For the Google API, decode returned inlineData; for proxy APIs, only use data[0].url from the provider the user configured.
3. Save the generated image to the resolved output path. If the API returns inline binary data, decode it locally instead of pasting it into chat.
4. Verify the file exists and is non-empty before reporting success.
5. For Chinese text edits, follow references/chinese-text.md.

Supported Models

• Google Official: gemini-2.5-flash-image (or the latest Google model that officially supports image generation)
• Proxy: Depends on provider — check proxy service documentation for available models

Error Handling

• No API Key: Report "missing config/secrets.md" and show setup instructions from secrets.example.md. Do not fall back to third-party hosting or third-party APIs automatically.
• API error 4xx/5xx: Display status code and error message.
• Network timeout: Retry once, then report failure.
• Wrong provider config: Validate API_PROVIDER is either google or proxy.
• Missing local image for edit mode: Stop and ask for a valid local path instead of silently switching to text-to-image.
• Technical diagram request: Redirect to the relevant diagram skill instead of forcing image generation.

Output Contract

Return:

• provider used (google or proxy)
• generation mode (text-to-image or image-to-image)
• output file path
• one-line note about any fallback, retry, or skipped feature

Do not expose API keys, raw base64 payloads, or full response bodies in chat.

Final Checklist

• Correct provider selected from config
• Prompt mode matches the request
• Output file exists locally
• Secrets were not echoed back to the user