Research map / Video / Video generation
Direct Autoregressive Diffusion Distillation via Error-aware Causal Pretraining
Jiaxing Li, Kaichen Huang, Baixin Xu, Zexiang Liu, Xianglong He, Zile Wang, Junyao Gao, Yang Liu, Ying He, Bo An, Yangguang Li
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| use Error Forcing instead of Teacher Forcing for pretraining (ablation) | 50-step causal model · VBench | VBench Total score: 79.15 -> 79.64 | Improved |
| use Error Forcing instead of Diffusion Forcing for pretraining (ablation) | 50-step causal model · VBench | VBench Total score: 78.22 -> 79.64 | Improved |
| initialize DMD from Error Forcing instead of Teacher Forcing (ablation) | 4-step student · VBench | VBench Total score: 80.77 -> 81.35 | Improved |
| initialize DMD from Error Forcing instead of Diffusion Forcing (ablation) | 4-step student · VBench | VBench Total score: 80.19 -> 81.35 | Improved |
| store errors frame-wise rather than chunk-wise (ablation) | Error Forcing · VBench | VBench Total score: 79.08 -> 79.64 | Improved |
| store errors frame-wise rather than window-wise (ablation) | Error Forcing · VBench | VBench Total score: 77.38 -> 79.64 | Improved |
| sample errors randomly rather than timestep-aware (ablation) | Error Forcing · VBench | VBench Total score: 79.32 -> 79.64 | Improved |
| inject errors into target latents as well as context (ablation) | Error Forcing · VBench | VBench Total score: 77.36 -> 79.64 | Improved |
| use a shared error instead of independent errors for context and target (ablation) | Error Forcing · VBench | VBench Total score: 77.40 -> 79.64 | Improved |
| add ODE distillation before DMD to a bidirectional init (ablation) | Bidirectional student · VBench | VBench Total score: 76.25 -> 80.79 | Improved |
| use error-aware causal init instead of bidirectional+ODE init (ablation) | DMD student · VBench | VBench Total score: 80.79 -> 81.35 | Improved |
| add causal ODE distillation after error-aware causal pretraining (ablation) | DMD student · VBench | VBench Total score: 81.35 -> 80.99 | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- Causal Forcing — Causal Forcing: Autoregressive Diffusion Distillation Done Right for High-Quality Real-Time Interactive Video Generation (ICML 2026) · better on 3 of 3 shared results
- Longlive — LongLive: Real-time Interactive Long Video Generation (ICLR 2026) · better on 2 of 2 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Video-Mirai: Autoregressive Video Diffusion Models Need Foresight (NeurIPS 2026 · similarity 0.91)
- Self Forcing: Bridging the Train-Test Gap in Autoregressive Video Diffusion (NeurIPS 2025 · similarity 0.91)
- AR-Diffusion: Asynchronous Video Generation with Auto-Regressive Diffusion (CVPR 2025 · similarity 0.91)
- Autoregressive Distillation of Diffusion Transformers (CVPR 2025 · similarity 0.90)
- From Slow Bidirectional to Fast Autoregressive Video Diffusion Models (CVPR 2025 · similarity 0.90)
- RAVEN: Real-time Autoregressive Video Extrapolation with Consistency-model GRPO (NeurIPS 2026 · similarity 0.90)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
