All-Purpose Image Banana

Problem It Solves

When general text-to-image tools cannot reliably render readable in-image text, keep consistency across multiple reference images, or match a request specifically for Nano Banana, a generic illustration skill may be insufficient. All-Purpose Image Banana reduces image generation to a fixed API contract: model and capability are constant, requests use POST /api/v1/llm/invoke, and results are fetched by polling GET /api/v1/llm/tasks/{task_id}.

How It Works

• Payment gate: If payment capability is missing, the flow stops; after a 402, retry with the same JSON body and add only the required payment headers.
• Async output: After submission, track task_id through pending, running, and succeeded, then read image URLs from data.result.data[].url.
• Prompt discipline: For text-to-image, structure the prompt around subject, scene, lighting, style, and composition; quoted in-image text should state the exact wording and placement.
• Reference editing: Pass b64 or url entries in images, declare what must be preserved, then describe only the requested change.

Boundaries

Best for posters, infographics, clear readable text, high-resolution visuals, and multi-reference consistency. General illustrations without special requirements may use a general drawing skill instead. Confirm the upstream model is enabled before calling, and verify text, identity, clothing, and composition after output.