---
license: apache-2.0
---

JiT-B with DINOv3

The model follows the description of the related paper (Figure 2) and the [loss calculation](https://github.com/kaist-cvml/PixelREPA/blob/main/denoiser_pixelREPA.py#L127) of the adapter model.

The inference is not compatible with the official codebase.

This repo is a continuation of the [portrait demo](https://huggingface.co/nebulette/segmentation-aware-jit-b).

Random weights are provided.

- 2603.14366