Search papers, labs, and topics across Lattice.
This paper introduces Adversarial Flow Distillation (AFD), a novel on-policy framework for distilling black-box video generators into autoregressive student models. AFD trains a Bradley-Terry discriminator to estimate teacher-student discrepancy based on prompt-paired queries and student rollouts, converting this advantage into flow-matching updates on the student's noised states. Experiments demonstrate that AFD enhances motion and physics fidelity in generated videos while maintaining overall quality, offering a practical approach for distilling proprietary or heterogeneous video generators.
Distilling black-box video generators into autoregressive models doesn't require teacher scores or complex alignment鈥攋ust cleverly paired rollouts and a discriminator.
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.
