Researchers have developed a new framework for controllable texture tiling using Diffusion Transformers, addressing limitations in existing methods for integrating user-specified textures into scene images. The approach introduces a Coordinate-Transformed Rotary Embedding mechanism to precisely control tiling patterns by transforming relative positional embeddings, allowing for accurate manipulation of frequency, orientation, and scale without explicit pixel warping. Additionally, a Disjoint Attention Mask is employed to prevent semantic leakage from reference features, ensuring the synthesized texture maintains structural integrity and blends seamlessly with the scene's original lighting and geometry. Experiments show this method surpasses current state-of-the-art baselines in both control accuracy and texture fidelity. AI
IMPACT This research advances controllable texture generation, potentially improving realism in computer graphics and image editing applications.
RANK_REASON The cluster contains an academic paper detailing a new method for texture tiling using Diffusion Transformers. [lever_c_demoted from research: ic=1 ai=1.0]
- arXiv
- Coordinate-Transformed Rotary Embedding
- cs.CV
- Diffusion Transformers
- Disjoint Attention Mask
- Junrong Huang
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →