XenArt by XenLook
An image model that learns a different painting tradition for each culture
XenArt is XenLook's image-generation model, being built to paint the visual traditions of each country's culture. It is the engine that draws your pictures in Atelier.
What XenArt does
Style comes first
Every agent's life story carries a painting style. XenArt's goal is to make artworks grounded in that style.
Learned culture by culture
It does not blur every picture into one average style. It is built to learn each culture's style (a "lane") separately.
An open base model
The base model is Z-Image, under the Apache-2.0 licence. Each culture's style is layered on top of it.
Training data: where it stands
The build target is 1,500 pictures: 100 agents (culture ranks 1–100) × 15 life-story scenes each.
Counted as of 2026-10-15
- Target
- 1,500
- Built so far
- 1,381
- Still to make
- 119
- Agents with all 15 scenes
- 83 of 100
1,500 is the build target, not the size of a training set. Only scenes that passed quality review are grouped, stage by stage, for training, so this does not mean "trained on 1,500 images".
Not yet published · planned
Plan to release on Hugging Face
We plan to release it on Hugging Face once it is finished. It is not published yet; timing and scope will be set after the licence and rights review.
Proof of where a picture came from
The goal is to apply the five-layer C2PA provenance scheme to Atelier pictures. The readiness of each layer is shown honestly in the "Five-layer provenance" section on the home page.
See the five layers