Research map / Video / Video generation
AR-CoPO: Align Autoregressive Video Generation with Contrastive Policy Optimization
Dailan He, Guanlin Feng, Xingtong Ge, Yi ZHANG, Bingqi Ma, Guanglu Song, Yu Liu, Hongsheng 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 |
|---|---|---|---|
| add semi-on-policy alignment | Self-Forcing · VBench | Total: 82.15 -> 82.45 | Improved |
| add merged on-policy and semi-on-policy LoRA (scale 0.8) | Self-Forcing · VideoAlign | Overall: 7.76 -> 8.22 | Improved |
| add merged on-policy and semi-on-policy LoRA (scale 0.8) | Self-Forcing · VBench | Total: 82.15 -> 82.17 | No clear effect |
| add on-policy AR-CoPO alone | Self-Forcing · VideoAlign | Overall: 7.76 -> 8.51 | Improved |
| add on-policy AR-CoPO alone | Self-Forcing · VBench | Total: 82.15 -> 81.99 | Hurt |
| on-policy training with TA-only reward (ablation) | AR-CoPO · VideoAlign | MQ: 1.68 -> 0.25 | Hurt |
| on-policy training with TA-only reward (ablation) | AR-CoPO · VBench | Total: 82.15 -> 79.26 | Hurt |
| on-policy training with TA-only reward (ablation) | AR-CoPO · VideoAlign | TA: 2.28 -> 2.63 | Improved |
| add ratio clipping (semi-on-policy vs fully off-policy) (ablation) | AR-CoPO · VBench | Total: 67.99 -> 82.45 | Improved |
| add ratio clipping (semi-on-policy vs fully off-policy) (ablation) | AR-CoPO · VideoAlign | Overall: 4.23 -> 7.61 | Improved |
| increase on-policy LoRA merge scale from 0 to 1.0 (ablation) | AR-CoPO merged · VideoAlign | Overall: 7.61 -> 8.33 | Improved |
| increase on-policy LoRA merge scale from 0 to 1.0 (ablation) | AR-CoPO merged · VBench | Total: 82.45 -> 81.99 | 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 4 of 5 shared results
- LongLive — LongLive: Real-time Interactive Long Video Generation (ICLR 2026) · better on 5 of 5 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
- Flash-GRPO: Efficient Alignment for Video Diffusion via One-Step Policy Optimization (ICML 2026 · similarity 0.88)
- RAVEN: Real-time Autoregressive Video Extrapolation with Consistency-model GRPO (NeurIPS 2026 · similarity 0.88)
- Joint Alignment and Distillation for Video Generation via Sample-Guided Distribution Matching (ECCV 2026 · similarity 0.88)
- Improving Video Generation with Human Feedback (NeurIPS 2025 · similarity 0.88)
- AR-Diffusion: Asynchronous Video Generation with Auto-Regressive Diffusion (CVPR 2025 · similarity 0.87)
- Flow-GRPO: Training Flow Matching Models via Online RL (NeurIPS 2025 · similarity 0.86)
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.
