Dataset Viewer
Auto-converted to Parquet Duplicate
Search is not available for this dataset
video
video
6.04
6.92
label
class label
4 classes
0humo-1.7b
1infinitetalk-fusionx
2longcat-avatar-1.5
3ltx-2.5

Character dialogue model comparison

AI-generated test clips comparing video models on one input: the same image of two characters, the same Telugu dialogue audio and the same seed, where each model supports it. The video and motion are synthetic.

Model Status Clip Size Length FPS Whole run on one A100 80 GB Settings
LTX-2.5 Chosen ltx-2.5/video.mp4 704x1280 6.04 s 24 8.0 min Two-stage audio-to-video, 30 + 3 steps, convolutional decoder
InfiniteTalk + FusionX Chosen infinitetalk-fusionx/video.mp4 704x1280 6.12 s 25 34.1 min 8 steps, multi-speaker
LongCat-Video-Avatar 1.5 Compared longcat-avatar-1.5/video.mp4 736x1248 6.92 s 25 19.9 min 8 steps, multi-speaker
HuMo 1.7B (light model) Compared humo-1.7b/video.mp4 480x832 6.12 s 25 17.3 min 40 steps (official 50), reduced resolution

Notes

  • Times are one run each, including model loading, so they are not a formal benchmark.
  • Settings differ between models, as listed above. HuMo ran at a lower resolution than the others.
  • LongCat and InfiniteTalk received a separate audio track per speaker. LTX-2.5 and HuMo received one mixed track.
  • Lip sync and visual quality have not been independently measured.
  • The input image and dialogue audio come from this YouTube Short by Little Wonder World: https://www.youtube.com/shorts/YAcOxHII1C8. Rights in that material stay with its owner.
  • Each model has its own license. See LTX-2, InfiniteTalk, LongCat-Video and HuMo.
Downloads last month
154