On-Policy Adversarial Flow Distillation for Autoregressive Video Generation

• URL: http://arxiv.org/abs/2605.26105v1
• Date: Mon, 25 May 2026 17:58:12 GMT
• Title: On-Policy Adversarial Flow Distillation for Autoregressive Video Generation
• Abstract summary: Adversarial Flow Distillation (AFD) is an on-policy framework for heterogeneous black-box video distillation.<n>AFD queries the teacher and rolls out the current student on the same prompts.<n>AFD provides dense velocity-field supervision while requiring no teacher scores, latents, denoising trajectories, step alignment, or reverse-chain reinforcement learning.
• Score: 34.70887671905755
• License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
• Abstract: Autoregressive video generators are attractive for streaming, long-horizon, and interactive applications, but distilling strong black-box teachers into causal students remains difficult. The student must learn under its own rollout distribution, whereas practical teachers may expose only prompt-conditioned completed videos and may differ in architecture, capacity, temporal design, and sampling schedule. This interface makes supervised fine-tuning off-policy, score-based distillation inapplicable, and direct adversarial imitation too sparse for denoising-time credit assignment. We propose Adversarial Flow Distillation (AFD), an on-policy framework for heterogeneous black-box video distillation. AFD queries the teacher and rolls out the current student on the same prompts, trains a prompt-paired Bradley-Terry discriminator to estimate clean-sample teacher-student discrepancy, and converts the resulting on-policy advantage into forward-process flow-matching updates on the student's own noised states. Thus, AFD provides dense velocity-field supervision while requiring no teacher scores, latents, denoising trajectories, step alignment, or reverse-chain reinforcement learning. Experiments across two causal AR student families show that AFD consistently improves motion- and physics-sensitive generation while preserving general video quality, and ablations validate the importance of adaptive on-policy feedback and forward-process credit assignment. The method requires only clean teacher videos and student rollouts, providing a practical route for distilling proprietary or heterogeneous video generators into efficient autoregressive students.

Related papers

• Mask Forcing: Improving Autoregressive Video Diffusion Distillation via Dual-Noise Masking Rollout [28.90854720290184]
  Recent methods distill pretrained bidirectional video diffusion models into causal AR students through Distribution Matching Distillation (DMD)<n>The generated videos often suffer from over-saturation and over-smoothing issues, resulting in limited visual quality and realism.<n>We propose Mask Forcing, a Dual-Noise Masking Rollout strategy that perturbs the AR student self-rollout to mitigate mode collapse.
  arXiv  Detail & Related papers  (2026-09-08T17:50:05Z)
• Context-Matched Distillation: Teacher Causality for Autoregressive Video Distillation [97.02922768457239]
  Context-Matched Distillation (CMD) is a causal DMD framework that aligns teacher supervision with the information available when each target is generated.<n>Prefix Corruption stabilizes training by perturbing unreliable prefixes produced early in training while preserving this target-context alignment.<n> Experiments demonstrate state-of-the-art aggregate performance among autoregressive methods on both short- and long-video benchmarks.
  arXiv  Detail & Related papers  (2026-08-13T15:51:31Z)
• OPSD-V: On-Policy Self-Distillation for Post-Training Few-Step Autoregressive Video Generators [68.69611976783243]
  OPSD-V is an on-policy self-distillation paradigm for post-training few-step autoregressive (AR) video diffusion models.<n>Key idea is to introduce real long-video data as temporal context during training and use it to provide dense trajectory-level supervision.<n>Experiments show consistent improvements in visual quality, motion dynamics, and VBenchLong scores.
  arXiv  Detail & Related papers  (2026-07-09T17:59:11Z)
• Adversarial Dual On-Policy Distillation from Expressive Teacher [53.63143943452487]
  Flow Matching teacher is learned from demonstrations and co-trained with a lightweight student.<n>The teacher provides two complementary signals on student rollouts.<n>The action channel supplies dense local targets at student-visited states, stabilizing exploitation.
  arXiv  Detail & Related papers  (2026-05-26T14:38:03Z)
• Backtracking When It Strays: Mitigating Dual Exposure Biases in LLM Reasoning Distillation [43.020181623750865]
  We propose Monitoring Trajectories and Backtracking when it strays (MOTAB)<n>MOTAB monitors the student's on-policy generation against an adaptive safety boundary.<n>Experiments on the LIMO-v2 and AceReason datasets demonstrate that MOTAB effectively alleviates the dual exposure biases.
  arXiv  Detail & Related papers  (2026-05-19T06:42:52Z)
• Reward-Aware Trajectory Shaping for Few-step Visual Generation [48.211558783057086]
  We propose textbfReward-Aware Trajectory Shaping (RATS), a lightweight framework for preference-aligned few-step generation.<n>RATS effectively transfers preference-relevant knowledge from high-step generators without incurring additional test-time computational overhead.<n> Experimental results demonstrate that RATS substantially improves the efficiency--quality trade-off in few-step visual generation.
  arXiv  Detail & Related papers  (2026-04-16T11:52:11Z)
• OVD: On-policy Verbal Distillation [47.727229201069555]
  On-policy Verbal Distillation (OVD) is a memory-efficient framework that replaces token-level probability matching with trajectory matching.<n>OVD dramatically reduces memory consumption while enabling on-policy distillation from teacher models with verbal feedback.
  arXiv  Detail & Related papers  (2026-01-29T16:48:14Z)
• BAgger: Backwards Aggregation for Mitigating Drift in Autoregressive Video Diffusion Models [50.986189632485285]
  We introduce Backwards Aggregation (BAgger), a self-supervised scheme that constructs corrective trajectories from the model's own rollouts.<n>Unlike prior approaches that rely on few-step distillation and distribution-matching losses, BAgger trains with standard score or flow matching objectives.<n>We instantiate BAgger on causal diffusion transformers and evaluate on text-to-video, video extension, and multi-prompt generation.
  arXiv  Detail & Related papers  (2025-12-12T23:02:02Z)
• Training-Free Motion Customization for Distilled Video Generators with Adaptive Test-Time Distillation [53.877572078307935]
  Distilled video generation models offer fast and efficient but struggle with motion customization when guided by reference videos.<n>We propose MotionEcho, a training-free test-time distillation framework that enables motion customization by leveraging diffusion teacher forcing.
  arXiv  Detail & Related papers  (2025-06-24T06:20:15Z)
• Seeing in the Dark: A Teacher-Student Framework for Dark Video Action Recognition via Knowledge Distillation and Contrastive Learning [2.772895608190934]
  ActLumos is a teacher-student framework that attains single-stream inference while retaining multistream level accuracy.<n>Under single-stream inference, the distilled student attains state-of-the-art accuracy of 96.92% (Top-1) on ARID V1.0, 88.27% on ARID V1.5, and 48.96% on Dark48.
  arXiv  Detail & Related papers  (2025-02-06T02:26:47Z)
• Object-Centric Diffusion for Efficient Video Editing [64.71639719352636]
  Diffusion-based video editing has reached impressive quality. Such solutions typically incur heavy memory and computational costs to generate temporally-coherent frames. We propose modifications that allow significant speed-ups whilst maintaining quality.
  arXiv  Detail & Related papers  (2024-01-11T08:36:15Z)
• Delta Distillation for Efficient Video Processing [68.81730245303591]
  We propose a novel knowledge distillation schema coined as Delta Distillation. We demonstrate that these temporal variations can be effectively distilled due to the temporal redundancies within video frames. As a by-product, delta distillation improves the temporal consistency of the teacher model.
  arXiv  Detail & Related papers  (2022-03-17T20:13:30Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.