GPT-image2 & Nano-Banana via Mid-Relay API

The Problem It Solves

When leveraging AI for image generation, developers or creators often seek the flexibility to use models from different providers, such as GPT-image2 or Nano-Banana (e.g., Gemini-based image models). However, interfacing directly with multiple official APIs creates management overhead: separate account sign-ups, juggling different API keys, handling disparate calling formats, and potentially incurring higher costs. A relay station (or API proxy) is a common solution to this problem, acting as an aggregation layer between the user and the models. But how to efficiently and securely call these models through a relay still requires a standardized tooling workflow.

How It Works

This skill encapsulates the complexity into a streamlined command-line workflow via a Python script named api_image.py. Its core capabilities and key steps are as follows:

1. Intelligent Configuration & Initialization
- Automatic Protocol Selection: Based on the user's chosen model type (e.g., GPT-image2 or Banana/Gemini), the script automatically determines whether to use the openai or gemini protocol, eliminating the need for manual specification.
- Secure Configuration Saving: During first-time use, an Agent guides the user to input the relay's Base URL, Model ID, and API Key. The script then saves this configuration securely to the local file ~/.api-image/config.json, preventing key leakage.
- Connectivity Verification: After configuration, the script automatically runs a low-cost test generation to ensure the settings are correct and the relay is reachable.

2. Core Generation Capabilities
- Text-to-Image: Generates a new image based solely on a text prompt. Configuration details do not need to be re-supplied for each call.
- Image-to-Image: Supports editing and generation based on a reference image. Notably, for the OpenAI protocol (e.g., GPT-image2), the script strictly adheres to the official requirement by calling the /v1/images/edits endpoint and actually uploading the reference image, ensuring full functionality.
- Flexible Response Handling: Processes both Base64 data and image URL responses from the relay, and automatically corrects the file extension (e.g., .png, .jpeg) based on the actual image bytes.

3. Key Operational Rules
- The script follows a clear configuration priority: Command Arguments > Environment Variables > Local Config File > Default Values.
- For timeouts or ambiguous responses, no automatic retry is performed to prevent unexpected costs.

Scope and Caveats

This skill is focused on providing a reliable, concise API calling channel, not a full-featured image generation platform. Please note the following when using it:

• Supported Scope:
• Supports basic text-to-image, single/multi-image reference generation, and multi-image editing under the OpenAI protocol.
• Input formats supported include PNG, JPEG, and WebP.

• Does NOT provide: Mask-based inpainting, conversational multi-turn generation state management, or platform-specific metadata processing.

• Known Limitations:

• The GPT-image2 model does not support transparent background generation; the script will reject the --background transparent parameter.

• Requires a runtime environment of Python 3.9 or higher and relies solely on Python's standard library, avoiding the need for complex third-party dependencies.

• Error Handling Recommendations:

• For 401/403 errors, first verify the validity of the API Key and account balance.
• For 404/405 errors, check if the Base URL is correct and if the relay has enabled the corresponding model protocol.
• For 429 (rate limiting) or 5xx (server errors), report the situation but should not retry consecutively.
• On timeout or connection interruption, the result status is unknown; avoid automatically resending the request.

In summary, this is a practical utility script aimed at simplifying the relay calling process. Its stability and privacy ultimately depend on the service quality of the relay provider chosen by the user.