Research map / Video / Video generation
FreeMem: Enhancing Consistency in Long Video Generation via Tuning-Free Memory
Jibin Peng, Di Lin, Zhecheng Xu, Haoran Lu, Ruonan Liu, Wuyuan Xie, Miaohui Wang, Lingyu Liang, Yi Wang, Qing Guo
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 noise memory (ablation) | FreeMem on FreeNoise · FreeNoise 64 frames ablation | sub first: 85.24 -> 93.92 | Improved |
| add token memory on top of noise memory (ablation) | FreeMem on FreeNoise · FreeNoise 64 frames ablation | sub first: 93.92 -> 94.68 | Improved |
| add attention memory on top of noise and token memory (ablation) | FreeMem on FreeNoise · FreeNoise 64 frames ablation | sub first: 94.68 -> 95.96 | Improved |
| add noise memory instead of FreeInit (ablation) | FreeMem on FreeNoise · FreeNoise 64 frames ablation | sub first: 91.65 -> 93.92 | Improved |
| use both rearrange steps in noise memory rather than only the second (ablation) | FreeMem noise memory · FreeNoise 64 frames ablation | sub first: 93.58 -> 93.92 | Improved |
| average similar tokens in fusion instead of replacing them (ablation) | FreeMem token memory · FreeNoise 64 frames ablation | sub first: 84.70 -> 86.10 | Improved |
| update token memory instead of only using initial memory (ablation) | FreeMem token memory · FreeNoise 64 frames ablation | sub pre: 96.41 -> 97.09 | Improved |
| use dynamic attention memory instead of concatenating first-frame key/value (ablation) | FreeMem attention memory · FreeNoise 64 frames ablation | sub first: 86.50 -> 87.68 | Improved |
| add FreeMem to FreeNoise | FreeNoise · VBench 64 frames | sub first: 85.24 -> 95.96 | Improved |
| add FreeMem to StreamingT2V | StreamingT2V · VBench 64 frames | sub first: 82.89 -> 85.26 | Improved |
| add FreeMem for 256 frames | StreamingT2V base · 256 frames | sub first: 52.96 -> 58.81 | Improved |
| add FreeMem to FIFO-Diffusion (compute cost) | FIFO-Diffusion · VBench 64 frames | inference time (s): 770 ± 5 -> 1180 ± 5 | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 4 of 4 on most shared results.
- StreamingT2V — StreamingT2V: Consistent, Dynamic, and Extendable Long Video Generation from Text (CVPR 2025) · better on 4 of 4 shared results
- FreeNoise — FreeNoise: Tuning-Free Longer Video Diffusion via Noise Rescheduling (ICLR 2024) · better on 3 of 3 shared results
- FreeLong — FreeLong: Training-Free Long Video Generation with SpectralBlend Temporal Attention (NeurIPS 2024) · 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
- FreeLong: Training-Free Long Video Generation with SpectralBlend Temporal Attention (NeurIPS 2024 · similarity 0.90)
- FreeNoise: Tuning-Free Longer Video Diffusion via Noise Rescheduling (ICLR 2024 · similarity 0.89)
- StreamingT2V: Consistent, Dynamic, and Extendable Long Video Generation from Text (CVPR 2025 · similarity 0.88)
- ByTheWay: Boost Your Text-to-Video Generation Model to Higher Quality in a Training-free Way (CVPR 2025 · similarity 0.88)
- T2V-Turbo: Breaking the Quality Bottleneck of Video Consistency Model with Mixed Reward Feedback (NeurIPS 2024 · similarity 0.87)
- Ouroboros-Diffusion: Exploring Consistent Content Generation in Tuning-free Long Video Diffusion (AAAI 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.
