This framework minimizes maximum mean discrepancy in frozen self-supervised video representation spaces in frozen self-supervised video representation spaces, using a hybrid Nystr\"om--Monte Carlo estimator to balance approximation bias and sampling variance.
Abstract
Few-step autoregressive video generation commonly relies on Distribution Matching Distillation (DMD), requiring a bidirectional diffusion teacher and an online fake-score model. We instead learn the rollout distribution directly from reference videos, eliminating both score models during post-training. Our framework minimizes maximum mean discrepancy (MMD) in frozen self-supervised video representation spaces, using a hybrid Nystr\"om--Monte Carlo estimator to balance approximation bias and sampling variance. Memory-efficient replay and gradient subsampling make this objective practical. Using the same architecture and initialization as Self-Forcing, our 1.3B model improves the VBench Total score from 83.80 to 84.64 while retaining 17 FPS. Removing auxiliary score models also enables 14B post-training on eight H200 GPUs. Beyond distillation, learning from reference videos enables the acquisition of new visual styles, semantic concepts, and spatial priors without a target-specific diffusion teacher.
Few-step autoregressive video generation enables efficient streaming synthesis, but errors introduced in early temporal blocks are reused as context and can propagate through subsequent rollouts, leading to detail degradation, structural drift, and unstable motion. Existing distribution matching distillation (DMD) prim...
Fang-Yu Lin, Xingtong Ge, Lu Zhu et al.· arXiv.org· 0 citations
Autoregressive video models can generate minute-long videos in real time, but they produce generic subjects from text rather than specific subjects from user-provided images. Existing customization methods either require costly per-subject optimization or use pretrained conditioning networks that jointly process all vi...
Yunseung Ok, Hyunsoo Kim, Minseo Kim et al.· 0 citations
Rollout-Marginal Distillation is introduced, which retains the generated history for AR prediction but scores each chunk independently against a chunk teacher, ensuring its quality correction is not compromised by an imperfect temporal context.
Chen-Jian Gao, Zhi-Hao Hu, Jian-Qi Ma et al.· 0 citations
ViRDM, a teacher- and critic-free video post-training recipe that turns three-network distillation into generator-only post-training, reducing GPU memory use and training time while improving video quality, is introduced.
Zi-Chong Meng, Chong-Jian Ge, C. Huang et al.· arXiv.org· 2 citations· ⚡2
DyMD, a DMD framework that adapts both teacher supervision and critic fitting to the evolving student with no auxiliary modules at inference is proposed, and a 14B teacher is distill into a four-step 1.3B student with no auxiliary modules at inference.
Hao-Jun Xu, Jie Huang, Xin Lu et al.· arXiv.org· 0 citations
Mask Forcing, a Dual-Noise Masking Rollout strategy that perturbs the AR student self-rollout to mitigate mode collapse induced by reverse-KL mode seeking, improves multiple AR video diffusion distillation methods with higher visual quality efficiently, without incorporating real video data or additional post-training...
Zhuo-Ran Zhao, Sheng-Ju Qian, Tong-Tong Liang et al.· 2 citations
Known for his clear and elegant writing style, Bertsekas shaped fields from control and optimization to large-scale computation and artificial intelligence.