Get Started
Home
Topics
Search
Library
Research questionHow can diffusion image and video generators be preference-aligned without inefficient training exploration or inference-time search?Policy-gradient preference optimization can explore inefficiently, become trapped in local optima, and exploit reward signals in ways that reduce semantic faithfulness or visual realism. Alignment must also avoid requiring extra search during generation.
AI
Alignment & Safety
Diffusion Models
Image Generation
Reinforcement Learning
Video Generation
Latest papersRecent research connected to this question, newest first.Step Back to Move Forward: Reflection-Aware Preference Optimization for Visual GenerationThe source concerns architecture-agnostic reinforcement-learning post-training for diffusion text-to-image and text-to-video generators. It reports experiments on both task types and states that its approach adds no inference-time overhead.research paper · Sep 3, 2026
Related questions
How can video generators align with human preferences while avoiding costly RL and post-distillation model collapse?How can text-to-image diffusion models preserve prompt alignment without the extra sampling cost of classifier-free guidance?How can text-to-image diffusion models use multiple candidates and continuous rewards beyond pairwise preferences?How can prompt embeddings be optimized during diffusion inference to balance aesthetics, prompt alignment, and resource use?