whatmodel.sh / video / HunyuanVideo
HunyuanVideo
↗ share
+ contribute
HunyuanVideo is an open-weights video model from Tencent, released December 2024. Its published size is 13B parameters. Licence: tencent-hunyuan-community.
$ how it compares
The video models ranked either side of HunyuanVideo, out of 139 with a score, then more from Tencent.
| # | model | score | where | ||
|---|---|---|---|---|---|
| 103 |  UniVideo | 39±3 | comparison only | ||
| 104 |  T2V-Turbo (MS) | 39±6 | comparison only | ||
| 105 | HunyuanVideo | 39±3 | api price unknown +local | ||
| 106 |  Pandora (world model) | 39±7 | comparison only | ||
| 107 | T2V-Turbo-v2 | 38±5 | api price unknown +local | ||
| more from Tencent | |||||
| 55 | HunyuanVideo 1.5 (I2V) | 53±2 | api $0.075/sec +local | ||
| 58 |  HY-OmniWeaving | 52±4 | comparison only | ||
| 59 | HunyuanVideo 1.5 | 52±2 | api $0.075/sec +local | ||
$ what it's good at
text to video▆▄▃▄▆▅█▃▂▂▂▄ #49 of 68 · 97% of its score · 6 tests
Trails LTX-2.3, T2V-Turbo-v2 here, though they rank lower overall.
| test | result | vs the others in its run | counts |
|---|---|---|---|
| VideoPhy-2 (human, joint SA & PC) | 17.2% ✎ | ▄···▄·▄·█··▄ beats 1/5 | 26% |
| SkyReels-Bench T2V (human rating) | 56.4% ✎ | ▄····▄···█·▄ beats 1/4 | 18% |
| VideoPhy (auto-evaluator average) | 35.0% ✎ | ▅·▃▅··▃█···▃ beats 4/9 | 15% |
| VideoVerse overall (Gemini 2.5 Pro judge) | 898.07 / 1608 ✎ | █·▄··▄·····█ beats 1/5 | 15% |
| VBench (T2V total) | 82.7% ✎ | █···▄▄▄····▄ beats 2/5 | 13% |
| VBench T2V total (long-prompt version, SkyReels-V2 runs) | 82.7% ✎ | █···█···█··█ beats 2/3 | 11% |
$ where to run it
Through an API
api · fal · price unknown · fal.ai ↗
Other ways to run it
version
- Run it as published · needs ~31.7 GB, NVIDIA only · from github.com/Tencent-Hunyuan
python3 sample_video.py --video-size 544 960 --video-length 129 --infer-steps 50 --prompt "A cat walks on the grass, realistic style." --flow-reverse --use-cpu-offload --save-path ./results
Official weights and documentation: huggingface.co/tencent ↗
$ how the score was worked out · VideoPhy / VideoPhy-2 officia… 41%, SkyReels-V2: Infinite-length… 29%, VideoVerse: Does Your T2V Gen… 15% 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
| source | results | share of the score |
|---|---|---|
 VideoPhy / VideoPhy-2 official repository leaderboards | 2 | 41% |
 SkyReels-V2: Infinite-length Film Generative Model | 2 | 29% |
| VideoVerse: Does Your T2V Generator Have World Model Capability to Synthesize Videos? | 1 | 15% |
| Training-Free Efficient Video Generation via Dynamic Token Carving | 1 | 13% |
| prior · pull toward the middle | 3% | |
| weighted average | = 39.0 |
$ published speed measurements · 1
Reference results for the hardware, engine and settings below. They aren't verified speeds of the downloads above or predictions for your machine.
| NVIDIA L20 (48GB; single GPU) | bf16 transformer, fp16 pipeline (per the guide's code); VAE tiling enabled | 3675.71 s/video | diffusers HunyuanVideoPipeline (ParaAttention guide baseline; versions not stated) · tencent/HunyuanVideo (revision refs/pr/18, diffusers format); 1280x720; 129 frames; 30 inference steps; 1 GPU; baseline without FBCache/context parallelism except the memory optimisation needed to avoid OOM (line 16: VAE tiling in the code) Baseline row; the prose (line 18) states 129 frames at 720p in 30 steps in 3675.71 s. Contributed to the diffusers docs by the ParaAttention author (PR #10544, Jan 2025); datacenter GPU, not a consumer card. | source ↗ |
$ improve this page
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