whatmodel.sh / video / ZeroScope v2 XL (576w → XL pipeline)

ZeroScope v2 XL (576w → XL pipeline)

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

ZeroScope v2 XL (576w → XL pipeline) is an open-weights video model from cerspense, released June 2023. Licence: cc-by-nc-4.0.

Catalogue identity notes

Table 1 June 2023 ZeroScope, 1024×576 output. Two-stage generation using 576w base then XL video-to-video upscaler. Separate from zeroscope-v2-576w.

Model information sources

· Weights licence ↗

23.5±2.3#92 of 93 video modelsfrom 4 benchmark results across 1 source

$ how it compares

The video models ranked either side of ZeroScope v2 XL (576w → XL pipeline), out of 139 with a score, then more from cerspense.

#modelscorewhereRAMreleased
135LVDM (text-to-video) HKUST / Tencent AI Lab25±3local~6.1GB2023-05
136ModelScope T2V 1.7B Alibaba DAMO Academy25±2local~4.6GB2023-03
137Floor33 Pictures (August 2023) Floor33 Pictures24±2comparison only2023-08
138ZeroScope v2 XL (576w → XL pipeline) cerspense24±2local~15GB2023-06
139CogVideo 9B (original) Tsinghua / Zhipu23±7local~52GB2022-05
more from cerspense
126ZeroScope v2 576w cerspense27±3local~7.9GB2023-06

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

$ what it's good at

text to video▆▄▃▄▆▅█▃▂▂▂▄ #67 of 68 · 98% of its score · 4 tests
testresultvs the others in its rundifficultycountswhy
EvalCrafter 2023 motion quality53.61 / 100 ✎▄····▄██··▄▄ beats 4/7very easy · avg 85.14 / 100▇▇▇▇▇▇▇▇▇▇ 25%at its level
EvalCrafter 2023 visual quality53.41 / 100 ✎▄█·····▄·█▄▄ beats 2/7very easy · avg 87.87 / 100▇▇▇▇▇▇▇▇▇▇ 25%at its level
EvalCrafter 2023 temporal consistency58.91 / 100 ✎▄···▄▄▄▄▄··█ beats 3/7very easy · avg 88.67 / 100▇▇▇▇▇▇▇▇▇▇ 24%at its level
EvalCrafter 2023 text-video alignment51.21 / 100 ✎▃··▂▂·····▂█ beats 0/7very easy · avg 87.86 / 100▇▇▇▇▇▇▇▇▇ 23%at its level

$ where to run it

Other ways to run it

  1. Run it as published · needs ~15.3 GB, NVIDIA only · from huggingface.co/cerspense (official)
    uv run --with torch --with diffusers --with transformers --with accelerate python -c "import torch; from PIL import Image; from diffusers import DiffusionPipeline, DPMSolverMultistepScheduler; from diffusers.utils import export_to_video; p=DiffusionPipeline.from_pretrained('cerspense/zeroscope_v2_576w', torch_dtype=torch.float16); p.scheduler=DPMSolverMultistepScheduler.from_config(p.scheduler.config); p.enable_model_cpu_offload(); p.enable_vae_slicing(); prompt='Darth Vader is surfing on waves'; f=p(prompt, num_inference_steps=40, height=320, width=576, num_frames=36).frames[0]; x=DiffusionPipeline.from_pretrained('cerspense/zeroscope_v2_XL', torch_dtype=torch.float16); x.scheduler=DPMSolverMultistepScheduler.from_config(x.scheduler.config); x.enable_model_cpu_offload(); x.enable_vae_slicing(); v=[Image.fromarray((fr * 255).round().astype('uint8')).resize((1024, 576)) for fr in f]; export_to_video(x(prompt, video=v, strength=0.6).frames[0], 'video_1024.mp4')"

Official weights and documentation: huggingface.co/cerspense ↗

$ how the score was worked out · EvalCrafter: original 2023 te… 98%

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
EvalCrafter: original 2023 text-to-video evaluations4▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇ 98%CC-BY-4.0
prior · pull toward the middle▇ 2%
weighted average= 23.5

$ 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