DistillAlign: Coordinating Mode Covering and Mode Seeking in Autoregressive Video Distillation

Jiaxing Li^1,2*, Kai Zou^1*, Cindy Zhou^1,3, Kaichen Huang^1,2, Junyao Gao¹, Zile Wang¹,
Yang Liu¹, Bin Liu¹, Bo An², Yangguang Li^1†

¹Riemann Dynamics    ²Nanyang Technological University    ³Wellington College, UK

_{^*Equal contribution    ^†Corresponding author}

Project Page · Paper · Code

DistillAlign aligns and balances the mode-covering and mode-seeking objectives of the multi-stage video distillation pipeline — using only a 1.3B DMD teacher, it already surpasses baselines refined with a 14B DMD teacher.

Checkpoints

All released generators are Wan2.1-1.3B students; the size in each name refers to the teacher used during training.

Model	Checkpoint	Description
Initializer (1.3B teacher)	distillalign_init_1p3b_teacher.pt	Pre-DMD initializer, trained toward a Wan2.1-T2V-1.3B teacher
Initializer (14B teacher)	distillalign_init_14b_teacher.pt	Pre-DMD initializer, trained toward a Wan2.1-T2V-14B teacher
Distilled (1.3B teacher)	distillalign_distill_1p3b_teacher.pt	Final joint-distilled generator, Wan2.1-T2V-1.3B DMD teacher
Distilled (14B teacher)	distillalign_distill_14b_teacher.pt	Final joint-distilled generator, Wan2.1-T2V-14B DMD teacher

Each checkpoint stores a plain {"generator": state_dict} and loads directly with the released inference and evaluation code.

Distillation Data

Prompts from VidProM, 25K clean-latent samples per teacher:

Data	Download	Description
Wan2.1-14B data	DistillAlign_14B_25K	25K distillation samples generated by Wan2.1-T2V-14B
Wan2.1-1.3B data	DistillAlign_1p3b_25K	25K distillation samples generated by Wan2.1-T2V-1.3B

Evaluation Caches

Teacher reference artifacts for the teacher-normalized distribution evaluation (16 prompts x 16 seeds, 25-step UniPC, timestep shift 8.0, CFG 5.0; V-JEPA2 ViT-H features, 2560-D):

Cache	Download	Description
Wan2.1-1.3B teacher reference features	wan2.1_t2v_1.3b_reference_vjepa2.npz	256 x 2560 features, ready for --teacher-features
Wan2.1-14B teacher reference features	wan2.1_t2v_14b_reference_vjepa2.npz	256 x 2560 features, ready for --teacher-features
Teacher reference videos	teacher_caches/videos	256 reference videos per teacher, for inspection and re-extraction

See teacher_caches/README.md for the exact protocol.

Citation

@misc{li2026distillalign,
  title         = {DistillAlign: Coordinating Mode Covering and Mode Seeking in Autoregressive Video Distillation},
  author        = {Li, Jiaxing and Zou, Kai and Zhou, Cindy and Huang, Kaichen and Gao, Junyao and Wang, Zile and Liu, Yang and Liu, Bin and An, Bo and Li, Yangguang},
  year          = {2026},
  eprint        = {2607.26811},
  archivePrefix = {arXiv},
  primaryClass  = {cs.CV},
  url           = {https://arxiv.org/abs/2607.26811}
}

License

The code and released weights follow the Apache License 2.0; upstream Wan2.1 model weights remain subject to their own licenses.