It handles the whole loop around an image — describe it, interrogate one aspect, compare it against another, read the raw file metadata — and caches what it has already looked at, so an iteration loop does not turn into an API bill.
An image analysis server built on OpenAI's vision models, running locally against files on your disk. Results are cached, so asking the same thing twice does not bill twice. It also generates and edits images from text prompts using DALL-E and GPT-Image-1.
- A full description of an image at a path, or an answer to the specific question you attach to it
- Targeted analysis of one aspect — objects, text, colours, composition or emotion — rather than a general description
- Two images compared, with the focus of the comparison set by you
- Technical metadata read off the file itself, for PNG, JPEG, GIF and WebP
- A cache you can inspect and clear, so repeated analysis is free and stale results are removable
Uv and Python; the server is launched from the repo directory. An OpenAI API key in the environment — the vision and generation calls bill to your OpenAI account. It reads images by path, so the files have to be on the machine running it.
Build from source — clone the repository and build it, then point your client at the binary