whatmodel.sh / video / LaVie (base T2V)

LaVie (base T2V)

+ contribute
Update model informationAdd a place to run itAdd a test resultPropose a benchmarkCorrect a benchmark scoreEdit data on GitHub

LaVie (base T2V) is an open-weights video model from Shanghai AI Lab (Vchitect), released September 2023. Licence: lavie-weights.

31.8±1.9#82 of 93 video modelsfrom 6 benchmark results across 4 sources

$ how it compares

The video models ranked either side of LaVie (base T2V), out of 139 with a score.

#modelscorewhereRAMreleased
120Pika 1.0 Pika32±3comparison only2023-11
121Open-Sora 1.1 HPC-AI Tech32±7local~14GB2024-04
122LaVie (base T2V) Shanghai AI Lab (Vchitect)32±2local~4.9GB2023-09
123LTX-Video 0.9 (2B) Lightricks32±3local~5.3GB2024-11
124Cosmos-1.0-Autoregressive-5B-Video2World NVIDIA31±5comparison only2025-01

see all video models · find the ones that fit your machine

$ what it's good at

text to video▆▄▃▄▆▅█▃▂▂▂▄ #56 of 68 · 99% of its score · 6 tests

Beats Open-Sora 1.1 here, though they rank higher overall.

testresultvs the others in its rundifficultycountswhy
GenAI-Arena text-to-video (anonymous Elo)964.0 ✎▃▃···▅▃·▃▅·█ beats 3/10avg 1229▇▇▇▇▇▇▇▇▇▇▇▇▇ 31%sharp
VideoPhy (human, SA=1 & PC=1)15.7% ✎▄▄▄·▄▄█····▄ beats 3/7medium · avg 48%▇▇▇▇▇▇▇▇▇ 23%sharp
AIGVE-60K human perception MOS (LOVE)49.3% ✎▄▄·█··█▄▄▄▄▄ beats 5/10easy · avg 73%▇▇▇▇▇▇▇▇ 19%at its level
AIGVE-60K human text-video correspondence MOS (LOVE)48.2% ✎▄▄·█▄█▄··▄▄▄ beats 4/10easy · avg 70%▇▇▇▇ 11%at its level
VideoPhy (auto-evaluator average)41.0% ✎▅·▃▅··▃█···▃ beats 6/9medium · avg 56%▇▇▇ 7.8%at its level
ChronoMagic-Bench-150 GPT4o-MTScore (time-lapse metamorphosis)50.0% ✎▄·▄█▄▄▄·▄·▄█ beats 3/10medium · avg 71%▇▇ 6.2%at its level

$ where to run it

Other ways to run it

  1. Run it as published · needs ~4.9 GB, NVIDIA only · from github.com/Vchitect
    git clone https://github.com/Vchitect/LaVie && cd LaVie && hf download Vchitect/LaVie lavie_base.pt --local-dir pretrained_models && hf download CompVis/stable-diffusion-v1-4 --local-dir pretrained_models/stable-diffusion-v1-4 && cd base && python pipelines/sample.py --config configs/sample.yaml

Official weights and documentation: huggingface.co/Vchitect ↗

$ how the score was worked out · GenAI-Arena (TIGER-Lab) text-… 31%, LOVE: Benchmarking and Evalua… 31%, VideoPhy / VideoPhy-2 officia… 30% and 1 more

Each result implies a score on its own, after adjusting for how hard its test is and who ran it. The score is their weighted average, with a small pull toward the middle for thinly tested models. Each test's weight is its counts column above; here it's added up by who ran the tests. How it works

sourceresultsshare of the scorelicence
GenAI-Arena (TIGER-Lab) text-to-video leaderboard1▇▇▇▇▇▇▇▇▇ 31%MIT
LOVE: Benchmarking and Evaluating Text-to-Video Generation and Video-to-Text Interpretation (AIGVE-60K)2▇▇▇▇▇▇▇▇▇ 31%CC-BY-4.0
VideoPhy / VideoPhy-2 official repository leaderboards2▇▇▇▇▇▇▇▇▇ 30%MIT
ChronoMagic-Bench: A Benchmark for Metamorphic Evaluation of Text-to-Time-lapse Video Generation1▇▇ 6%CC-BY-4.0
prior · pull toward the middle▇ 1%
weighted average= 31.8

$ improve this page

Update model informationAdd a place to run itAdd a test resultPropose a benchmarkCorrect a benchmark scoreEdit data on GitHub

Use a short GitHub form or edit the data through a pull request. Every number needs a source; here's which ones we can use.

data updated · sources and licences · json