Research map / Video / Video generation
SteinsGate: Adding Causality to Diffusions for Long Video Generation via Path Integral
Yufei Huang, Liangyu Yuan, Changxi Chi, Yunfan Liu, Cheng Tan, Siyuan Li, Jingbo Zhou, Haitao Lin, Chang Yu, Stan Z. 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 Video Path Integral to I2V-AR continuation (ablation) | SteinsGate (Wan2.1) · InstructVC benchmark (multi-action long video) | CSCV: 0.74 -> 0.82 | Improved |
| add Video Path Integral to I2V-AR continuation (ablation) | SteinsGate (Wan2.1) · InstructVC benchmark (multi-action long video) | motion smoothness: 0.94 -> 0.97 | Improved |
| add Guidance Interval (VPI only in high-noise stage) (ablation) | SteinsGate · InstructVC benchmark (multi-action long video) | CSCV: 0.81 -> 0.82 | No clear effect |
| add History-aligned Redistribution (ablation) | SteinsGate · InstructVC benchmark (multi-action long video) | CSCV: 0.79 -> 0.82 | Improved |
| add Path Convergence Guidance (ablation) | SteinsGate · InstructVC benchmark (multi-action long video) | CSCV: 0.78 -> 0.82 | Improved |
| use SteinsGate instead of other training-free methods (DiTCtrl, FIFO) | inference-time long video generation · InstructVC benchmark (multi-action long video) | CSCV: 0.76 -> 0.82 (vs DiTCtrl); 0.71 -> 0.82 (vs FIFO) | Improved |
| use SteinsGate (training-free) vs training-based diffusion forcing (SkyReel-V2, MAGI-1) | SteinsGate · InstructVC benchmark (multi-action long video) | CSCV: 0.83 / 0.82 vs 0.82 | No clear effect |
| replace Temporal Action Binding with concatenated prompts and equal durations (ablation) | SteinsGate · InstructVC benchmark (multi-action long video) | action completeness/order | Improved |
| raise PCG weight from 1.0 to 1.5 (extrapolation) (ablation) | Path Convergence Guidance · InstructVC benchmark (multi-action long video) | CSCV: 0.78 -> 0.82 | Improved |
| raise selected history frames K from 1 to 3 (ablation) | Video Path Integral · InstructVC benchmark (multi-action long video) | CSCV: 0.74 -> 0.82 | Improved |
| increase history ratio from 20% to 30% of segment length (ablation) | SteinsGate · InstructVC benchmark (multi-action long video) | CSCV: 0.82 -> 0.79 | Hurt |
| add Video Path Integral to I2V-AR | SteinsGate · inference cost 15 s video | generation time (min): 30 -> 38 | 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.
- DiTCtrl — DiTCtrl: Exploring Attention Control in Multi-Modal Diffusion Transformer for Tuning-Free Multi-Prompt Longer Video Generation (CVPR 2025) · better on 3 of 3 shared results
- FIFO-Diffusion — FIFO-Diffusion: Generating Infinite Videos from Text without Training (NeurIPS 2024) · better on 3 of 3 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 0.89)
- MovieDreamer: Hierarchical Generation for Coherent Long Visual Sequences (ICLR 2025 · similarity 0.89)
- StreamingT2V: Consistent, Dynamic, and Extendable Long Video Generation from Text (CVPR 2025 · similarity 0.88)
- FlowMo: Variance-Based Flow Guidance for Coherent Motion in Video Generation (NeurIPS 2025 · similarity 0.88)
- LongScape: Advancing Long-Horizon Embodied World Models with Context-Aware MoE (NeurIPS 2026 · similarity 0.88)
- CineOrchestra: Unified Entity-Centric Conditioning for Cinematic Video Generation (NeurIPS 2026 · similarity 0.88)
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.
