Research map / Video / Video generation
LEO: Generative Latent Image Animator for Human Video Synthesis
Yaohui Wang, Xin Ma, Xinyuan Chen, Cunjian Chen, Antitza Dantcheva, Bo Dai, Yu Qiao
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 Linear Motion Condition (LMC) to LMDM (ablation) | LEO · TaichiHD | FVD: 118.6 -> 100.4 | Improved |
| add Linear Motion Condition (LMC) to LMDM (ablation) | LEO · FaceForensics | FVD: 60.03 -> 52.32 | Improved |
| add Transition Diffusion Model for long TaichiHD videos (ablation) | LEO · TaichiHD long video | repeated motion occurrence (%): 0.45 -> 0.02 | Improved |
| condition on a real starting image instead of cDDPM-generated one | LEO · TaichiHD128 | FVD16: 100.4 -> 57.6 | Improved |
| condition on a real starting image | LEO · FaceForensics | FVD16: 52.3 -> 35.9 | Improved |
| compare LEO to DIGAN at 256x256 resolution | LEO (uncond) · TaichiHD256 | FVD16: 156.7 -> 122.7 | Improved |
| generate 128-frame videos with LEO instead of TATS | LEO · TaichiHD128 long | FVD128: 1194.58 -> 155.54 | Improved |
| generate 128-frame videos with LEO instead of StyleGAN-V | LEO · FaceForensics long | FVD128: 89.34 -> 96.28 | Hurt |
| unconditional LEO versus StyleGAN-V on FaceForensics short videos | LEO (uncond) · FaceForensics | FVD16: 47.4 -> 52.3 | Hurt |
| apply LEO to general videos versus large diffusion models | LEO · UCF101 | FVD16: 367.23 -> 1356.2 | Hurt |
| apply LEO instead of StyleGAN-V on UCF101 | LEO · UCF101 | FVD16: 1431.0 -> 1356.2 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- StyleGAN-V — StyleGAN-V: A Continuous Video Generator With the Price, Image Quality and Perks of StyleGAN2 (CVPR 2022) · better on 3 of 4 shared results
- MoStGAN-V — MoStGAN-V: Video Generation With Temporal Motion Styles (CVPR 2023) · better on 1 of 1 shared result
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
- VideoJAM: Joint Appearance-Motion Representations for Enhanced Motion Generation in Video Models (ICML 2025 · similarity 0.89)
- EverAnimate: Minute-Scale Human Animation via Latent Flow Restoration (NeurIPS 2026 · similarity 0.88)
- Make Pixels Dance: High-Dynamic Video Generation (CVPR 2024 · similarity 0.88)
- FlowMo: Variance-Based Flow Guidance for Coherent Motion in Video Generation (NeurIPS 2025 · similarity 0.88)
- MagicAnimate: Temporally Consistent Human Image Animation using Diffusion Model (CVPR 2024 · similarity 0.88)
- Enhanced Fine-Grained Motion Diffusion for Text-Driven Human Motion Synthesis (AAAI 2024 · 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.
