whatmodel.sh / video / Cosmos-1.0-Diffusion-7B-Text2World

Cosmos-1.0-Diffusion-7B-Text2World

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

Cosmos-1.0-Diffusion-7B-Text2World is an open-weights video model from NVIDIA, released January 2025. Its published size is 7B parameters. Licence: nvidia-open-model.

Model information sources

Release date ↗ · Weights licence ↗ · Parameter count ↗

NVIDIA's Cosmos technical report (maker paper). 7B = diffusion transformer only; T5 text encoder, CV8x8x8 tokenizer, 12B prompt upsampler and guardrails are separate.

45.9±3.1#61 of 93 video modelsfrom 3 benchmark results across 2 sources

$ how it compares

The video models ranked either side of Cosmos-1.0-Diffusion-7B-Text2World, out of 139 with a score, then more from NVIDIA.

#modelscorewhereRAMreleased
84Cosmos-Predict2.5-14B NVIDIA47±3comparison only2025-12
85CogVideoX-5B-I2V Zhipu AI (THUDM)46±4local~13GB2024-09
86Cosmos-1.0-Diffusion-7B-Text2World NVIDIA46±3local~30GB2025-01
87Hailuo 01 (MiniMax video-01 / I2V-01) MiniMax46±4api $0.083/sec2024-08
88Kling 1.6 (Standard mode) Kuaishou (Kling AI)45±5comparison only2024-12
more from NVIDIA
38Cosmos3-Super NVIDIA58±4local~160GB2026-05
60Cosmos3-Nano NVIDIA52±3local~42GB2026-05
72Cosmos-1.0-Diffusion-7B-Video2World NVIDIA49±5comparison only2025-01

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

$ what it's good at

text to video▆▄▃▄▆▅█▃▂▂▂▄ #39 of 68 · 96% of its score · 3 tests

Beats Wan 2.2 TI2V 5B here, though they rank higher overall.

testresultvs the others in its rundifficultycountswhy
VideoPhy-2 (human, joint SA & PC)24.1% ✎▄···▄·▄·█··▄ beats 3/5hard · avg 33%▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇ 52%at its level
Cosmos 3D consistency: camera-pose estimation success rate on generated videos lab62.6% ✎█·█·······██ beats 2/3easy · avg 68%▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇ 42%sharp · at its level
Cosmos 3D consistency: Sampson epipolar error on generated videos (lower is better) lab0.3550 ✎█···█···█··█ beats 3/3avg 0.385▇ 2.0%blunt · noisy

$ where to run it

Other ways to run it

  1. Run it as published · needs ~29.8 GB, NVIDIA only · from huggingface.co/nvidia (official)
    uv run --with torch --with diffusers --with transformers --with accelerate --with imageio-ffmpeg --with cosmos_guardrail python -c "import torch; from diffusers import CosmosTextToWorldPipeline; from diffusers.utils import export_to_video; p=CosmosTextToWorldPipeline.from_pretrained('nvidia/Cosmos-1.0-Diffusion-7B-Text2World', torch_dtype=torch.bfloat16); p.to('cuda'); export_to_video(p(prompt='A sleek, humanoid robot stands in a vast warehouse filled with neatly stacked cardboard boxes on industrial shelves.').frames[0], 'output.mp4', fps=30)"

Official weights and documentation: huggingface.co/nvidia ↗

$ how the score was worked out · VideoPhy / VideoPhy-2 officia… 52%, Cosmos World Foundation Model… 44%

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
VideoPhy / VideoPhy-2 official repository leaderboards1▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇ 52%MIT
Cosmos World Foundation Model Platform for Physical AI (arXiv:2501.03575v3) · lab-reported2▇▇▇▇▇▇▇▇▇▇▇▇▇ 44%CC-BY-4.0
prior · pull toward the middle▇ 4%
weighted average= 45.9

$ 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