Research map / Video / Video generation
Turbo2K: Towards Ultra-Efficient and High-Quality 2K Video Synthesis
Jingjing Ren, Wenbo Li, Zhongdao Wang, Haoze Sun, Bangzhen Liu, Haoyu Chen, Jiaqi Xu, Aoxue Li, Shifeng Zhang, Bin Shao, Yong Guo, Lei Zhu
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 |
|---|---|---|---|
| synchronize teacher and student timesteps in feature distillation instead of data fine-tuning (ablation) | LTX-Video student · Distillation ablation | MUSIQ: 60.36 -> 64.73 | Improved |
| synchronize teacher and student timesteps instead of fixing teacher t=0.1 (ablation) | LTX-Video student · Distillation ablation | MUSIQ: 61.12 -> 64.73 | Improved |
| add data fine-tuning to basic LTX (ablation) | LTX-Video · Distillation ablation | MUSIQ: 58.80 -> 60.36 | No clear effect |
| synchronized-timestep distillation vs basic LTX on semantic score (ablation) | LTX-Video · Distillation ablation | Seman.: 22.47 -> 24.48 | Improved |
| use late-step LR features (t=0.1) instead of t=t_hr guidance (ablation) | Turbo2K two-stage · Guidance ablation | Seman.: 18.53 -> 24.60 | Improved |
| use feature guidance (t=0.1) instead of LR latent guidance (ablation) | Turbo2K two-stage · Guidance ablation | MUSIQ: 49.42 -> 52.98 | Improved |
| use feature guidance (t=0.1) instead of LR latent guidance (ablation) | Turbo2K two-stage · Guidance ablation | NIQE: 5.04 -> 4.03 | Improved |
| compare Turbo2K 4B against CogVideoX-2B at 49 frames 2K | Turbo2K · 2K inference time | s/step: 94.68 -> 5.08 | Improved |
| Turbo2K vs CogVideoX-5B VBench total | Turbo2K · VBench-Long | Total Score: 81.61 -> 82.78 | Improved |
| Turbo2K vs HunyuanVideo VBench total | Turbo2K · VBench-Long | Total Score: 83.24 -> 82.78 | Hurt |
| Turbo2K vs RealBasicVSR as upscaler | Turbo2K · VSR comparison | MUSIQ: 51.48 -> 52.45 | 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.
- CogVideoX-5B — CogVideoX: Text-to-Video Diffusion Models with An Expert Transformer (ICLR 2025) · better on 5 of 7 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
- TurboVSR: Fantastic Video Upscalers and Where to Find Them (ICCV 2025 · similarity 0.88)
- Fast-Vid2Vid++: Spatial-Temporal Distillation for Real-Time Video-to-Video Synthesis (TPAMI 2024 · similarity 0.88)
- Training-Free Efficient Video Generation via Dynamic Token Carving (NeurIPS 2025 · similarity 0.87)
- Encapsulated Composition of Text-to-Image and Text-to-Video Models for High-Quality Video Synthesis (CVPR 2025 · similarity 0.86)
- T2V-Turbo: Breaking the Quality Bottleneck of Video Consistency Model with Mixed Reward Feedback (NeurIPS 2024 · similarity 0.85)
- LinGen: Towards High-Resolution Minute-Length Text-to-Video Generation with Linear Computational Complexity (CVPR 2025 · similarity 0.85)
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.
