smartapi-generate-image

SmartAPI Image Generation

Workflow

1. Keep the active Codex reasoning model on a text-capable model. Never select gpt-image-2 as the Codex reasoning model.
2. Turn the user's visual request into a concrete generation prompt without changing requested subjects, text, branding, format, or composition.
3. Call the generate_image MCP tool once. It spends SmartAPI balance.
4. Use size=auto unless the user requests dimensions or orientation. Pass requested WIDTHxHEIGHT values unchanged; SmartAPI reports the normalized effective size.
5. Use quality=high unless the user explicitly asks for another quality.
6. Use output_format=png unless the user requests JPEG or WebP.
7. After success, inspect the saved local file with the available image-viewing tool.
8. Return the image to the user and state its effective size, format, charged tokens, and saved path.

Safety

• Treat the generation as a paid external action.
• Never place SMARTAPI_API_KEY in prompts, tool arguments, source files, logs, or replies.
• Never retry a timeout, cancellation, network failure, or ambiguous server failure automatically. Explain that the billing outcome may be ambiguous and ask before a new generation.
• Do not request n>1, streaming, URL output, or image edits; the current SmartAPI endpoint supports one non-streaming b64_json generation.
• Do not copy or move the generated file outside its configured output directory unless the user asks and the normal filesystem policy permits it.
• Do not claim 4K output. Report the exact effective_size returned by SmartAPI.

Failure Handling

• Authentication errors: tell the user the SmartAPI key is missing or invalid without echoing it.
• Insufficient balance or token limit: report the server error; do not fall back to another account or endpoint.
• Existing filename: choose a different filename only when it is semantically equivalent and does not overwrite user data.
• Malformed image data: do not save or display it.