AINeutralarXiv – CS AI · 6h ago6/10
🧠
FreeStyle: Free Control of Style-Content Dual-Reference Generation from Community LoRA Mining
FreeStyle introduces a scalable framework for dual-reference image generation that synthesizes images preserving content structure while adopting separate style references, addressing the challenge of style-content separation through community LoRA mining and novel disentanglement mechanisms. The approach tackles a critical bottleneck in large-scale triplet dataset availability and achieves improved balance between style alignment, content preservation, and leakage suppression compared to existing methods.