whatmodel.sh / video / ModelScope T2V 1.7B
ModelScope T2V 1.7B
↗ share
+ contribute
ModelScope T2V 1.7B is an open-weights video model from Alibaba DAMO Academy, released March 2023. Its published size is 1.7B parameters. Licence: cc-by-nc-4.0.
$ how it compares
The video models ranked either side of ModelScope T2V 1.7B, out of 139 with a score.
| # | model | score | where |
|---|---|---|---|
| 134 |  Show-1 | 25±2 | local |
| 135 |  LVDM (text-to-video) | 25±3 | local |
| 136 | ModelScope T2V 1.7B | 25±2 | local |
| 137 |  Floor33 Pictures (August 2023) | 24±2 | comparison only |
| 138 | ZeroScope v2 XL (576w → XL pipeline) | 24±2 | local |
$ what it's good at
text to video▆▄▃▄▆▅█▃▂▂▂▄ #65 of 68 · 99% of its score · 9 tests
Beats LVDM (text-to-video) here, though they rank higher overall.
| test | result | vs the others in its run | counts |
|---|---|---|---|
| GenAI-Arena text-to-video (anonymous Elo) | 830.4 ✎ | ▃▃···▅▃·▃▅·█ beats 0/10 | 23% |
| AIGVE-60K human perception MOS (LOVE) | 38.0% ✎ | ▄▄·█··█▄▄▄▄▄ beats 1/10 | 13% |
| EvalCrafter 2023 motion quality | 52.47 / 100 ✎ | ▄····▄██··▄▄ beats 1/7 | 12% |
| EvalCrafter 2023 visual quality | 53.09 / 100 ✎ | ▄█·····▄·█▄▄ beats 1/7 | 12% |
| EvalCrafter 2023 temporal consistency | 57.80 / 100 ✎ | ▄···▄▄▄▄▄··█ beats 2/7 | 12% |
| EvalCrafter 2023 text-video alignment | 54.46 / 100 ✎ | ▂··▃▂·····▂█ beats 1/7 | 11% |
| AIGVE-60K human text-video correspondence MOS (LOVE) | 43.7% ✎ | ▄▄·█▄█▄··▄▄▄ beats 1/10 | 7.8% |
| FETV manual video quality (mean of static and temporal quality) | 3.93 / 5 ✎ | █·····█····█ beats 2/2 | 4.3% |
1 more test
| test | result | vs the others in its run | counts |
|---|---|---|---|
| ChronoMagic-Bench-150 GPT4o-MTScore (time-lapse metamorphosis) | 60.2% ✎ | ▄·▄█▄▄▄·▄·▄█ beats 8/10 | 4.2% |
$ where to run it
Other ways to run it
- Run it as published · needs ~4.6 GB, NVIDIA only · from huggingface.co/ali-vilab
python -c "import torch;from diffusers import DiffusionPipeline;from diffusers.utils import export_to_video;p=DiffusionPipeline.from_pretrained('ali-vilab/text-to-video-ms-1.7b',torch_dtype=torch.float16,variant='fp16');p.enable_model_cpu_offload();export_to_video(p('Spiderman is surfing',num_inference_steps=25).frames[0],'modelscope.mp4')"
Weights and documentation: huggingface.co/ali-vilab ↗
$ how the score was worked out · EvalCrafter: original 2023 te… 47%, GenAI-Arena (TIGER-Lab) text-… 23%, LOVE: Benchmarking and Evalua… 21% and 2 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 |
|---|---|---|
 EvalCrafter: original 2023 text-to-video evaluations | 4 | 47% |
 GenAI-Arena (TIGER-Lab) text-to-video leaderboard | 1 | 23% |
| LOVE: Benchmarking and Evaluating Text-to-Video Generation and Video-to-Text Interpretation (AIGVE-60K) | 2 | 21% |
| FETV: A Benchmark for Fine-Grained Evaluation of Open-Domain Text-to-Video Generation (arXiv:2311.01813v3) | 1 | 4% |
| ChronoMagic-Bench: A Benchmark for Metamorphic Evaluation of Text-to-Time-lapse Video Generation | 1 | 4% |
| prior · pull toward the middle | 1% | |
| weighted average | = 24.5 |
$ 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