Research map / Video / Video generation
Lightning Fast Caching-based Parallel Denoising Prediction for Accelerating Talking Head Generation
Jianzhi Long, Wenhao Sun, Rong-Cheng Tu, Dacheng Tao
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 input latents estimation to parallel denoising prediction (ablation) | LightningCP on EchoMimic · HDTF subset | FVD: 460.21 -> 225.16 | Improved |
| add input latents estimation to parallel denoising prediction (ablation) | LightningCP on Hallo · HDTF subset | FVD: 208.46 -> 200.33 | Improved |
| add input latents estimation (ablation) | LightningCP on EchoMimic · HDTF subset | E-FID: 0.226 -> 0.233 | No clear effect |
| add Decoupled Foreground Attention on top of LCP (ablation) | LightningCP on Hallo · Hallo FLOPs | FLOPs (x10^12/clip): 745.72 -> 679.56 | Improved |
| add Decoupled Foreground Attention on top of LCP (ablation) | LightningCP on Hallo · Hallo speedup | speedup: 2.94 -> 3.08 | Improved |
| add reference feature removal (ablation) | LCP + DFA on Hallo · MEAD | Sync: 6.19 -> 6.43 | Improved |
| add reference feature removal (ablation) | LCP + DFA on Hallo · Hallo speedup | speedup: 3.08 -> 3.15 | Improved |
| add DFA on top of LCP (ablation) | LightningCP on Hallo · HDTF | FID: 4.99 -> 5.10 | No clear effect |
| use LightningCP + DFA + rm instead of FasterDiffusion | talking head diffusion models · HDTF | FVD: 140.56 -> 121.80 (Hallo) | Improved |
| use LightningCP + DFA + rm instead of DeepCache | Hallo · Hallo speedup | speedup: 1.84 -> 3.15 | Improved |
| fewer diffusion steps limits non-key timesteps | MEMO and EchoMimic Acc. · Speedup | speedup: 3.15 (Hallo) vs 2.33 (MEMO) vs 1.45 (EchoMimic Acc.) | Hurt |
Closest papers by meaning
- Communication-Efficient Diffusion Denoising Parallelization via Reuse-then-Predict Mechanism (NeurIPS 2025 · similarity 0.86)
- Long-Term TalkingFace Generation via Motion-Prior Conditional Diffusion Model (ICML 2025 · similarity 0.86)
- VideoMLA: Low-Rank Latent KV Cache for Minute-Scale Autoregressive Video Diffusion (NeurIPS 2026 · similarity 0.86)
- FasterCache: Training-Free Video Diffusion Model Acceleration with High Quality (ICLR 2025 · similarity 0.86)
- MoDiTalker: Motion-Disentangled Diffusion Model for High-Fidelity Talking Head Generation (AAAI 2025 · similarity 0.85)
- FIFO-Diffusion: Generating Infinite Videos from Text without Training (NeurIPS 2024 · 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.
