Stay Fair! Ensuring Group Fairness in Diffusion Models Across Guidance Scales
Decouples fairness from guidance scale in diffusion models by equalizing guidance distributions or offsetting null embeddings
12 days ago
Decouples fairness from guidance scale in diffusion models by equalizing guidance distributions or offsetting null embeddings
BiDPO jointly optimizes image and text preferences with region-guided alignment to boost compositional fidelity in T2I generation.
Boosts text-to-video efficiency by combining hybrid full-sparse attention with Skiparse-2D and Sparse Sequence Parallelism to cut communication and maintain quality.
Directs narrative-aware video generation by conditioning on multiple keyframes, enabling single-shot, multi-shot, and extension scenarios
Unifies layers in a 20B masked region diffusion model for multi-layer transparent image generation and editing
Enables temporally coherent video generation with decentralized training, doubling FVD improvement and boosting CLIP similarity and aesthetics under equal compute.
