Simple API endpoint for image alt description generation, powered by OpenAI Vision.

This is inspired by Ice Cubes. It's really cool.

• Written in Deno.
• Serverless, on Deno Deploy.
• There's no front-end.
• Uses OpenAI Vision.

Requires Deno.

Add environment variable OPENAI_API_KEY. Could create a .env file and put the variable there too.

• deno task dev - Run the server with watch mode, for development.
• deno task debug - Run the server with watch mode and debugging, for development.
• deno task start - Run the server.

Additional environment variables:

• OPENAI_BASE_URL - Custom OpenAI-compatible API base URL (e.g. https://api.openai.com/v1). Defaults to OpenAI's official endpoint.
• OPENAI_MODEL - Model to use, defaults to gpt-5.4-nano.
• OPENAI_API_TYPE - API style to use: responses (default, OpenAI Responses API) or chat (older /v1/chat/completions compatible with most other providers).
• MAX_TOKENS - Maximum output tokens, defaults to 140.
• UPLOAD_LIMIT - Maximum uploaded image size in bytes, defaults to 10485760 (10MB).

Most OpenAI-compatible providers only support the older /v1/chat/completions endpoint. To switch to one:

1. Set OPENAI_API_TYPE=chat.
2. Set OPENAI_BASE_URL to the provider's OpenAI-compatible base URL.
3. Set OPENAI_MODEL to the provider's model name.
4. Set OPENAI_API_KEY to the provider's API key.

Gemini supports OpenAI-compatible requests via the Gemini OpenAI compatibility endpoint.

OPENAI_API_TYPE=chat \
OPENAI_BASE_URL=https://generativelanguage.googleapis.com/v1beta/openai/ \
OPENAI_MODEL=gemini-2.5-flash \
OPENAI_API_KEY=your-gemini-api-key \
MAX_TOKENS=1024 \
deno task start

• GET /?image=<image-url>
• POST / with image key as multipart/form-data.

Response:

{
  "description": "a picture of a cat"
}

Error response:

{
  "error": "Failed to generate description"
}

MIT.

• https://github.com/Dimillian/IceCubesApp/
• https://github.com/JeremiahDMoore/gpt4-vision