Research map / Video / Video generation
Helios: Real Real-Time Long Video Generation Model
Shenghai Yuan, Yuanyang Yin, Zongjian Li, Xinwei Huang, Xiao Yang, Li Yuan
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 Guidance Attention (ablation) | Helios-Base · Long video ablation (Total) | Total score: 6.23 -> 6.47 | Improved |
| add causal mask on self-attention in Guidance Attention (ablation) | Helios-Base · Long video ablation (Total) | Total score: unstable training process | Hurt |
| add First Frame Anchor (ablation) | Helios-Base · Long video ablation (Total) | Total score: 5.51 -> 6.47 | Improved |
| add Frame-Aware Corrupt (ablation) | Helios-Base · Long video ablation (Total) | Total score: 4.70 -> 6.47 | Improved |
| add Coarse-to-Fine Learning (ablation) | Helios-Distilled · Long video ablation (Total) | Total score: 5.31 -> 6.34 | Improved |
| add Adversarial Post-Training (ablation) | Helios-Distilled · Long video ablation (Total) | Total score: 6.31 -> 6.34 | Improved |
| replace Self-Forcing distillation with Helios distillation (ablation) | Helios-Distilled · Long video ablation (Total) | Total score: 6.11 -> 6.34 | Improved |
| use unidirectional rather than bidirectional teacher (ablation) | Helios-Distilled · Long video ablation (Total) | Total score: 4.75 -> 6.34 | Improved |
| replace DMD with Decouple DMD (ablation) | Helios-Distilled · Long video ablation (Total) | Total score: 6.34 -> 5.21 | Hurt |
| add reward-weighted regression post-training (ablation) | Helios-Distilled · Long video ablation (Total) | Total score: 6.34 -> 6.23 | Hurt |
| add Flash Normalization and Flash RoPE (ablation) | Wan-2.1-T2V-14B · DiT inference time | seconds: 98.68 -> 84.41 | Improved |
| distill Helios-Base into Helios-Distilled | Helios-Base · 120-1440-frame long video (Total) | Total score: 6.57 -> 6.94 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 3 of 3 on most shared results.
- LongLive — LongLive: Real-time Interactive Long Video Generation (ICLR 2026) · better on 3 of 3 shared results
- Reward Forcing — Reward Forcing: Efficient Streaming Video Generation with Rewarded Distribution Matching Distillation (CVPR 2026) · better on 2 of 2 shared results
- Rolling Forcing — Rolling Forcing: Autoregressive Long Video Diffusion in Real Time (ICLR 2026) · better on 1 of 1 shared result
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
- From Slow Bidirectional to Fast Autoregressive Video Diffusion Models (CVPR 2025 · similarity 0.88)
- Grid Diffusion Models for Text-to-Video Generation (CVPR 2024 · similarity 0.88)
- VideoMAR: Autoregressive Video Generation with Continuous Tokens (NeurIPS 2025 · similarity 0.87)
- FreeLong: Training-Free Long Video Generation with SpectralBlend Temporal Attention (NeurIPS 2024 · similarity 0.87)
- SnapGen-V: Generating a Five-Second Video within Five Seconds on a Mobile Device (CVPR 2025 · similarity 0.87)
- LongDiff: Training-Free Long Video Generation in One Go (CVPR 2025 · similarity 0.87)
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.
