Cactus Compute lists Mirage, a vision model for image understanding and structured extraction, on Hugging Face

Cactus Compute has published Mirage, a vision model for image understanding and structured extraction, on Hugging Face under the Cactus-Compute organization. The model page lists 12,844 downloads in the last month.

What is listed

The Hugging Face repository tags Mirage as Image-Text-to-Text. It provides three items: mirage.cact, mirage.safetensors and the model config. The page links to model details and a playground, a Python API, runtime binaries and supported platforms, and source code.

How to run it

Mirage is used through the cactus-compute Python package:

pip install -U cactus-compute

import cactus
print(cactus.Mirage().describe("photo.png"))

The package downloads and caches the engine and weights on first use.

Ecosystem on Hugging Face

The model tree shows one finetune of Mirage, and one Space uses the model. Mirage is part of a collection titled "Automation Foundation Models For Tiny Devices," which contains 6 items.

No Inference Provider currently deploys the model, according to the page. Hugging Face offers a prompt to ask for provider support.

What this means

Mirage is distributed as a local package with its own engine and weights rather than through a hosted API, since no Inference Provider serves it. Its collection title points to small-device deployment as the context, and the stated focus on structured extraction positions it for turning images into usable data rather than only captioning them. The 12,844 monthly downloads indicate early uptake of the repository.