---
license: apache-2.0
pipeline_tag: image-to-image
---

Color gelbooru images

A mini Gemma 4 unified model, it colors the input image based on the character tags.

![](images/preview.png)

Keywords
- pixel space model
- U-shaped transformer blocks
- cross-attention between encoder and decoder blocks
- 2D RoPE with center at origin
- position embedding with both cropped and non-square images

Source data
- realistic render (similar to academic art movement)
- noripro
- vspo!

References
- 2204.00631
- 2508.02324
- 2511.13720
- 2603.15031