Siming Fu

TEMPFLOW-GRPO: WHEN TIMING MATTERS FOR GRPO IN FLOW MODELS

SAIL: Self-Amplified Iterative Learning for Diffusion Model Alignment with Minimal Human Feedback

Consis-GCPO: Consistency-Preserving Group Causal Preference Optimization for Vision Customization

MOSAIC: Multi-Subject Personalized Generation via Correspondence-Aware Alignment and Disentanglement

LLaVA-MoD: Making LLaVA Tiny via MoE-Knowledge Distillation

MS-Diffusion: Multi-subject Zero-shot Image Personalization with Layout Guidance