Reading pictures with Iconclass

A 9-billion-parameter vision model that classifies artworks with Iconclass codes.

Iconclass is the standard iconographic classification used to catalogue what a picture depicts. Upload an artwork (or pick an example) and the model proposes codes, decoded to plain English. Optionally set how many codes you want, or paste codes your catalogue already has and ask what’s missing.

Try an example

Cataloguing depth

model decides·2·3·4·5·6·8·10codes

Already catalogued

The model will only suggest codes for aspects not yet covered. Press Enter to apply.

The artwork

Drop an image here
or click to choose a file

Predicted Iconclass codes

Awaiting an image…

raw model output

The artwork

Is this in the picture?

The model says this picture shows the following. Do you see it?

yes, it’s shown·no·can’t tellsaved

Model: davanstrien/qwen35-9b-iconclass-sft-multitask-2ep, a Qwen3.5-9B vision-language model fine-tuned on three tasks: classify an artwork, classify with a requested number of codes, and extend an existing partial catalogue record. Running on ZeroGPU. Code meanings decoded with the iconclass Python package. Predictions are a research artifact and may be incomplete or wrong — verify against iconclass.org.