Research map / Video / Video generation
Self-Forcing++: Towards Minute-Scale High-Quality Video Generation
Justin Cui, Jie Wu, Ming Li, Tao Yang, Xiaojie Li, Rui Wang, Andrew Bai, Yuanhao Ban, Cho-Jui Hsieh
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 |
|---|---|---|---|
| train with Self-Forcing++ (backward noise init, extended rollout) instead of Self-Forcing | Wan2.1-1.3B distilled AR video generator · 50s VBench Long + Gemini visual stability | visual stability: 40.12 -> 90.94 | Improved |
| Self-Forcing++ at 100s vs Self-Forcing | same · 100s dynamic degree | dynamic degree: 26.41 -> 54.12 | Improved |
| Self-Forcing++ on short 5s clips | same · VBench 5s | total score: 83.00 -> 83.11 | No clear effect |
| shorten attention window in training to 9 latent frames (ablation) | Self-Forcing · 50s visual stability | visual stability: 40.12 -> 52.50 | Mixed |
| add GRPO with optical-flow reward (ablation) | Self-Forcing++ · long video temporal consistency | optical-flow spikes | Improved |
| scale training budget from 1x to 25x (ablation) | Self-Forcing++ · long video generation length | coherent length | Improved |
| autoregressive KV-cache generation (Self-Forcing++) vs overlapped recomputation methods | Self-Forcing++ · temporal repetition | NoRepeat score: 67.19 (NOVA) / 73.44 (MAGI-1) vs 98.44 | Improved |
Closest papers by meaning
- FreeLong: Training-Free Long Video Generation with SpectralBlend Temporal Attention (NeurIPS 2024 · similarity 0.91)
- Self Forcing: Bridging the Train-Test Gap in Autoregressive Video Diffusion (NeurIPS 2025 · similarity 0.91)
- From Slow Bidirectional to Fast Autoregressive Video Diffusion Models (CVPR 2025 · similarity 0.91)
- Taming Teacher Forcing for Masked Autoregressive Video Generation (CVPR 2025 · similarity 0.90)
- AR-Diffusion: Asynchronous Video Generation with Auto-Regressive Diffusion (CVPR 2025 · similarity 0.90)
- Snap Video: Scaled Spatiotemporal Transformers for Text-to-Video Synthesis (CVPR 2024 · 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.
