XenArt LoRA v1.4
stylizing generations into fine art
2 min readupdated
This new LoRA (a lightweight add-on for a base model) for Z-Image (Tongyi-MAI/Z-Image) turns standard photorealistic outputs into hand-painted museum artwork. Version 1.4 was trained on 1,409 public-domain images (CC0 and Public Domain Mark) from five major museums and covers 11 distinct genres, ranging from Japanese Nihonga to European Symbolism. Compared to the previous release, the creator stripped out unwanted artifacts like frames and 족자 (traditional hanging scrolls) in East Asian styles. It is a solid tool for artists, researchers, and generative art hobbyists looking to create stylized illustrations.
Key features
- 11 detailed artistic styles, including Egyptian wall paintings, Nordic Romanticism, and South Asian miniature art.
- Fully open dataset: curated from the open-access collections of the Metropolitan Museum of Art, the Art Institute of Chicago, and other galleries.
- Cleaner outputs: v1.4 eliminates the random scroll borders and watermarks that occasionally ruined shots in v1.3.
- No trigger words required: styles respond directly to standard descriptive terms in your prompt.
What you can do
- Create atmospheric illustrations in the style of European Symbolism or American narrative painting for personal projects.
- Generate artwork styled after ancient frescoes or East Asian ink drawings.
- Experiment with style blending in
ComfyUIusing theZ-Imagebase model.
Example
Sample prompt for the Norwegian Romanticism style:
Nordic national-romantic painting, a lineman raising a utility pole on a stormy island shore, visible brushstrokes, painted canvas texture, wordless painted scene, clean unmarked surfacesRead more
Details• Adapter size: 170 MB (BF16, rank 32, 480 tensors).
• Base model:
Z-Image (Apache-2.0, not included).• Limitations: garbled text can occasionally appear on fine details like clocks or signs, and certain styles may inherit hand and facial anatomy quirks from the base model.










Comments