Research map / Video / Video generation
Long Context Tuning for Video Generation
Yuwei Guo, Ceyuan Yang, Ziyan Yang, Zhibei Ma, Zhijie Lin, Zhenheng Yang, Dahua Lin, Lu Jiang
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 Long Context Tuning (ablation) | pre-trained single-shot 3B model · VBench single-shot | Aesthetic: 52.09 -> 55.91 | Improved |
| add Long Context Tuning (ablation) | pre-trained single-shot 3B model · VBench single-shot | Quality: 63.37 -> 64.23 | Improved |
| add Long Context Tuning (ablation) | pre-trained single-shot 3B model · VBench single-shot | Text alignment: 30.81 -> 32.22 | Improved |
| add Long Context Tuning (ablation) | pre-trained single-shot 3B model · VBench single-shot | Consistency: 94.98 -> 95.06 | No clear effect |
| use auto-regressive instead of joint generation (bidirectional) (ablation) | LCT bidirectional · multi-shot VBench | Text alignment: 29.33 -> 30.24 | Improved |
| use auto-regressive instead of joint generation (bidirectional) (ablation) | LCT bidirectional · multi-shot VBench | Quality: 57.68 -> 56.62 | Hurt |
| use context-causal attention instead of bidirectional (auto-regressive) (ablation) | LCT · multi-shot VBench | Quality: 56.62 -> 50.31 | Hurt |
| use context-causal attention instead of bidirectional (auto-regressive) (ablation) | LCT · multi-shot | fidelity to history | Improved |
| raise history conditioning noise timestep (ablation) | LCT auto-regressive · multi-shot | error accumulation vs history fidelity | Mixed |
| add 1K steps of context-causal fine-tuning from bidirectional weights (ablation) | LCT context-causal · multi-shot | identity consistency | Improved |
| use LCT instead of VideoStudio | VideoStudio · scene-level | Text alignment: 28.00 -> 30.14 | Improved |
| use LCT instead of VideoStudio | VideoStudio · scene-level | User study: 2.14 -> 3.79 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- VideoStudio — VideoStudio: Generating Consistent-Content and Multi-Scene Videos (ECCV 2024) · better on 3 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
- Tune-A-Video: One-Shot Tuning of Image Diffusion Models for Text-to-Video Generation (ICCV 2023 · similarity 0.89)
- StoryDiffusion: Consistent Self-Attention for Long-Range Image and Video Generation (NeurIPS 2024 · similarity 0.88)
- StreamingT2V: Consistent, Dynamic, and Extendable Long Video Generation from Text (CVPR 2025 · similarity 0.88)
- ShotAdapter: Text-to-Multi-Shot Video Generation with Diffusion Models (CVPR 2025 · similarity 0.87)
- FreeLong: Training-Free Long Video Generation with SpectralBlend Temporal Attention (NeurIPS 2024 · similarity 0.87)
- FreeNoise: Tuning-Free Longer Video Diffusion via Noise Rescheduling (ICLR 2024 · 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.
