DealMaTe: Multi-Dimensional Material Transfer via Diffusion Transformer

Researchers introduced DealMaTe, a diffusion-transformer framework for material transfer that uses depth, normal, and lighting inputs instead of text guidance or reference networks.

It injects multi-dimensional 3D information through a lightweight Multi-Dim 3D Shader LoRA and optimizes attention with Shader Causal Mutual Attention and KV caching.

The approach aims to produce higher-fidelity material transfers with lower complexity and faster inference.

DealMaTe is designed to be more compatible across different objects and lighting conditions while simplifying diffusion-based image editing.