Research map / Video / Video generation
DLFR-Gen: Diffusion-based Video Generation with Dynamic Latent Frame Rate
Zhihang Yuan, Rui Xie, Yuzhang Shang, Hanling Zhang, Siyuan Wang, Shengen Yan, Guohao Dai, Yu Wang
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 |
|---|---|---|---|
| apply DLFR-Gen dynamic latent frame rate to HunyuanVideo | HunyuanVideo · VBench prompt subset | Latency (s): 714.90 -> 245.12 | Improved |
| apply DLFR-Gen with strictest threshold theta=0.9, k=15 | HunyuanVideo · VBench prompt subset | Speedup: 1.24 | Improved |
| lower similarity threshold theta from 0.9 to 0.5 (k=5) (ablation) | DLFR-Gen · VBench prompt subset | Speedup: 1.31 -> 2.92 | Improved |
| lower similarity threshold theta from 0.9 to 0.5 (k=5) (ablation) | DLFR-Gen · VBench prompt subset | SSIM: 0.631 -> 0.513 | Hurt |
| start compression later (k=5 to k=15, theta=0.5) (ablation) | DLFR-Gen · VBench prompt subset | Speedup: 2.92 -> 1.96 | Hurt |
| start compression later (k=5 to k=15, theta=0.5) (ablation) | DLFR-Gen · VBench prompt subset | VQA aesthetic score: 97.85 -> 96.57 | Hurt |
| start compression later (k=5 to k=15, theta=0.9) (ablation) | DLFR-Gen · VBench prompt subset | SSIM: 0.631 -> 0.723 | Improved |
| DLFR-Gen versus original on temporal flicker (k=5, theta=0.5) | HunyuanVideo · VBench prompt subset | Temporal flicker: 0.9793 -> 0.9919 | Improved |
| DLFR-Gen versus original on VQA (k=5, theta=0.5) | HunyuanVideo · VBench prompt subset | VQA aesthetic score: 98.63 -> 97.85 | Hurt |
| add the DLFR-Gen compression module (versus operating directly in latent space) (ablation) | DLFR-Gen · qualitative cat video (Fig. 4) | visual quality | Improved |
| add one-step denoise and renoise around the compression module (ablation) | DLFR-Gen · qualitative cat video (Fig. 4) | visual quality | Improved |
| start compression at earlier denoising timesteps (ablation) | DLFR-Gen · qualitative (Fig. 4) | visual quality | Hurt |
Closest papers by meaning
- Efficient Video Diffusion Models via Content-Frame Motion-Latent Decomposition (ICLR 2024 · similarity 0.91)
- FreeLong: Training-Free Long Video Generation with SpectralBlend Temporal Attention (NeurIPS 2024 · similarity 0.89)
- From Slow Bidirectional to Fast Autoregressive Video Diffusion Models (CVPR 2025 · similarity 0.89)
- Training-Free Efficient Video Generation via Dynamic Token Carving (NeurIPS 2025 · similarity 0.89)
- VideoTetris: Towards Compositional Text-to-Video Generation (NeurIPS 2024 · similarity 0.89)
- Ouroboros-Diffusion: Exploring Consistent Content Generation in Tuning-free Long Video Diffusion (AAAI 2025 · 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.
