SHIFT: Vision-Language Models for Accessible CH
Collection
Fine-tuned multimodal LLMs auto-generates rich cultural-heritage descriptions for visually impaired audiences in Romanian, Serbian and German. • 4 items • Updated • 1
This model is a LoRA fine-tune of Qwen/Qwen2.5-VL-3B-Instruct for museum-grade, accessibility-first descriptions in Romanian.
Model: JoseferEins/ArtQwen-Curator-RO
How to use: Follow the same instructions as in the DE and Multilingual (ML) model cards.
For a quick example, reuse the DE repo’s tools/run.py, sample image, and metadata format; just switch the LoRA ID and use a Romanian prompt (optionally a *_meta_ro.txt).
python tools/run.py \
--image botero.jpg \
--mode both \
--lora_id JoseferEins/ArtQwen-Curator-RO \
--metadata @botero_meta_ro.txt \
--prompt "Descrie doar ceea ce este vizibil, clar și accesibil. Folosește metadatele pentru titlu/artist/tehnică; nu inventa."
## Acknowledgements
This work was developed within the [SHIFT project](https://shift-europe.eu/) on cultural-heritage accessibility. It has received funding from the European Union’s Horizon Europe programme.
We thank the SHIFT consortium partners and museum collaborators for their feedback on accessibility guidelines and evaluation rubrics, and the blind and low-vision advisors whose input shaped the prompting and output style.
**Disclaimer:** Views and opinions expressed are those of the authors and do not necessarily reflect those of the European Union or the granting authority. Neither the European Union nor the granting authority can be held responsible for them.
Base model
Qwen/Qwen2.5-VL-3B-Instruct