Research map / Video / Video generation
RhymeFlow: Training Free Acceleration for Video Generation with Asynchronous Denoising Flow Scheduling
Chensheng Dai, Shengjun Zhang, Yifan Li, Zhang Zhang, Zheng Zhu, Yueqi Duan
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 content-aware keyframe selection (vs uniform) (ablation) | RhymeFlow · Wan 2.1 ablation (T2V 1.3B) | SSIM: 0.643 -> 0.783 | Improved |
| add content-aware keyframe selection (vs uniform) (ablation) | RhymeFlow · Wan 2.1 ablation (T2V 1.3B) | LPIPS: 0.183 -> 0.168 | Improved |
| add content-aware keyframe selection (vs uniform) (ablation) | RhymeFlow · Wan 2.1 ablation (T2V 1.3B) | PSNR: 24.293 -> 26.291 | Improved |
| add progressive asynchronous scheduling (vs fixed skip) (ablation) | RhymeFlow · Wan 2.1 ablation (T2V 1.3B) | PSNR: 25.399 -> 26.291 | Improved |
| add progressive asynchronous scheduling (vs fixed skip) (ablation) | RhymeFlow · Wan 2.1 ablation (T2V 1.3B) | Speedup: 1.40× -> 1.53× | Improved |
| add latent trajectory projection (ablation) | RhymeFlow · Wan 2.1 ablation (T2V 1.3B) | PSNR: 20.630 -> 26.291 | Improved |
| add latent trajectory projection (ablation) | RhymeFlow · Wan 2.1 ablation (T2V 1.3B) | Speedup: 1.60× -> 1.53× | Hurt |
| add rolling KV-cache management (ablation) | RhymeFlow · Wan 2.1 ablation (T2V 1.3B) | Peak memory (GB): 49.9 -> 42.6 | Improved |
| increase keyframes M from 3 to 5 at Tw=8 | RhymeFlow · Wan 2.1 (T2V 1.3B, 81 frames) | PSNR: 23.742 -> 27.707 | Improved |
| increase keyframes M from 3 to 5 at Tw=8 | RhymeFlow · Wan 2.1 (T2V 1.3B, 81 frames) | Speedup: 1.60× -> 1.39× | Hurt |
| increase warm-up steps Tw from 6 to 10 at M=4 | RhymeFlow · Wan 2.1 (T2V 1.3B, 81 frames) | PSNR: 22.588 -> 26.636 | Improved |
| add latent trajectory projection (latent error analysis) (ablation) | RhymeFlow · Wan 2.1 (T2V 1.3B, 81 frames) | PSNR: 24.44 -> 27.70 | Improved |
Closest papers by meaning
- FIFO-Diffusion: Generating Infinite Videos from Text without Training (NeurIPS 2024 · similarity 0.91)
- Equilibrium Forcing: Adaptive Video Generation Without Noise Conditioning (NeurIPS 2026 · similarity 0.90)
- Training-Free Efficient Video Generation via Dynamic Token Carving (NeurIPS 2025 · similarity 0.90)
- AdaDiff: Adaptive Step Selection for Fast Diffusion Models (AAAI 2025 · similarity 0.90)
- FreeNoise: Tuning-Free Longer Video Diffusion via Noise Rescheduling (ICLR 2024 · similarity 0.90)
- DSAQuant: Denoising-Stage-Aligned Quantization-Aware Training for Video Generation (NeurIPS 2026 · similarity 0.89)
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.
