Research map / Video / Video generation
Enhancing Train-Free Infinite-Frame Generation for Consistent Long Videos
Xiaokun Feng, Jiashu Zhu, Meiqi Wu, Chubin Chen, Fangyuan Mao, Haiyang Guo, Jiahong Wu, Xiangxiang Chu, Kaiqi Huang
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 two-stage training-inference alignment (TTA) to FIFO-Diffusion (ablation) | FIFO-Diffusion on VideoCrafter2 · VBench-Long | Overall Score: 95.02 -> 97.05 | Improved |
| add dual consistency enhancement (DCE) to FIFO-Diffusion (ablation) | FIFO-Diffusion on VideoCrafter2 · VBench-Long | Overall Score: 95.02 -> 96.75 | Improved |
| combine TTA and DCE (ablation) | FIFO-Diffusion on VideoCrafter2 · VBench-Long | Overall Score: 97.05 -> 97.82 | Improved |
| add zigzag iterative denoising (stage 1) (ablation) | FIFO-Diffusion on VideoCrafter2 · VBench-Long | Overall Score: 95.02 -> 96.78 | Improved |
| add unified-noise-level denoising (stage 2) after stage 1 (ablation) | MIGA on VideoCrafter2 · VBench-Long | Overall Score: 96.78 -> 97.05 | Improved |
| run only stage 2 directly on N noise latents (omit stage 1) (ablation) | MIGA · VBench-Long | Overall Score | Hurt |
| increase zigzag width Lzig from 1 to 4 (ablation) | MIGA on VideoCrafter2 · VBench-Long | Overall Score: 95.80 -> 97.00 | Improved |
| increase zigzag width Lzig beyond 4 (to 6 and 8) (ablation) | MIGA on VideoCrafter2 · VBench-Long | Overall Score: 97.00 -> 96.94 (Lzig=6), 96.99 (Lzig=8) | No clear effect |
| add long-range frame guidance latents (mguid 0 to 6) (ablation) | MIGA on VideoCrafter2 · VBench-Long | Overall Score: 95.80 -> 96.87 | Improved |
| increase guidance latents from 6 to 8 (ablation) | MIGA on VideoCrafter2 · VBench-Long | Overall Score: 96.87 -> 96.77 | Hurt |
| lower self-reflection adjustment threshold (more searches) (ablation) | MIGA · VBench-Long | Overall Score | Improved |
| lower the threshold further below 0.01 (ablation) | MIGA · VBench-Long | Overall Score / success rate | No clear effect |
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.
- FIFO-Diffusion — FIFO-Diffusion: Generating Infinite Videos from Text without Training (NeurIPS 2024) · better on 6 of 6 shared results
- FreePCA — FreePCA: Integrating Consistency Information across Long-short Frames in Training-free Long Video Generation via Principal Component Analysis (CVPR 2025) · better on 1 of 1 shared result
- FreeLong — FreeLong: Training-Free Long Video Generation with SpectralBlend Temporal Attention (NeurIPS 2024) · 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
- FreeLong: Training-Free Long Video Generation with SpectralBlend Temporal Attention (NeurIPS 2024 · similarity 0.88)
- LongDiff: Training-Free Long Video Generation in One Go (CVPR 2025 · similarity 0.88)
- Ouroboros-Diffusion: Exploring Consistent Content Generation in Tuning-free Long Video Diffusion (AAAI 2025 · similarity 0.87)
- FreePCA: Integrating Consistency Information across Long-short Frames in Training-free Long Video Generation via Principal Component Analysis (CVPR 2025 · similarity 0.87)
- FreeNoise: Tuning-Free Longer Video Diffusion via Noise Rescheduling (ICLR 2024 · similarity 0.87)
- StreamingT2V: Consistent, Dynamic, and Extendable Long Video Generation from Text (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.
