Research map / Video / Video generation
FlashLips: 100-FPS Mask-Free Latent Lip-Sync using Reconstruction Instead of Diffusion or GANs
Andreas Zinonos, Michał Stypułkowski, Antoni Bigata, Stavros Petridis, Maja Pantic, Nikita Drobyshev
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 |
|---|---|---|---|
| increase reference lip latents from 1 to 4 (ablation) | FlashLips Transformer · cross-audio | FVD: 41.38 -> 29.40 | Improved |
| increase reference lip latents from 4 to 32 (ablation) | FlashLips Transformer · cross-audio | LipScore: 0.37 -> 0.32 | Hurt |
| increase reference lip latents from 4 to 32 (ablation) | FlashLips Transformer · cross-audio | FVD: 29.40 -> 25.16 | Improved |
| increase reference lip latents from 1 to 4 (ablation) | FlashLips U-Net · cross-audio | FVD: 42.78 -> 33.57 | Improved |
| raise expression-encoder lips vector from 2D to 8D (ablation) | V1 lips encoder · reconstruction | PSNR: 27.11 -> 31.63 | Improved |
| raise V1 lips vector from 8D to 16D (ablation) | V1 lips encoder · reconstruction | PSNR: 31.63 -> 31.95 | Improved |
| add 4D lips-crop residual (V2) to 8D encoder (ablation) | lips encoder · reconstruction | PSNR: 31.63 -> 32.86 | Improved |
| add 4D lips-crop residual (V2) to 8D encoder (ablation) | lips encoder · cross-audio | ID similarity: 0.83 -> 0.80 | Hurt |
| enlarge residual from 4D to 16D (ablation) | lips encoder · cross-audio | ID similarity: 0.80 -> 0.63 | Hurt |
| use U-Net backbone instead of Transformer (ablation) | FlashLips Stage 1 · speed vs quality | FPS: 66.84 -> 109.41 FPS; FVD 12.31 -> 15.20 | Mixed |
| use FlashLips (reconstruction-only, one-step) instead of diffusion KeySync | lip-sync · reconstruction | FVD: 24.80 -> 12.31 | Improved |
| use FlashLips instead of LatentSync on pixel metrics | lip-sync · reconstruction | PSNR: 33.61 -> 32.86 | Hurt |
Closest papers by meaning
- OmniSync: Towards Universal Lip Synchronization via Diffusion Transformers (NeurIPS 2025 · similarity 0.91)
- SyncTalkFace: Talking Face Generation with Precise Lip-Syncing via Audio-Lip Memory (AAAI 2022 · similarity 0.89)
- StyleLipSync: Style-based Personalized Lip-sync Video Generation (ICCV 2023 · similarity 0.89)
- PointTalk: Audio-Driven Dynamic Lip Point Cloud for 3D Gaussian-based Talking Head Synthesis (AAAI 2025 · similarity 0.87)
- Perceptually Accurate 3D Talking Head Generation: New Definitions, Speech-Mesh Representation, and Evaluation Metrics (CVPR 2025 · similarity 0.86)
- Lips Are Lying: Spotting the Temporal Inconsistency between Audio and Visual in Lip-Syncing DeepFakes (NeurIPS 2024 · similarity 0.86)
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.
