whatmodel.sh / image / UniWorld-V1

UniWorld-V1

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

UniWorld-V1 is an open-weights image model from PKU-YuanGroup, released June 2025. Its published size is 20.37B parameters.

Model information sources

Release date ↗ · Parameter count ↗

HF Hub safetensors total for the official repo (pinned revision): 20,365,253,696 F32 across its 17 model shards (the card does not itemise components or state a size). Extra .bin/.pt files (SigLIP redux projector, denoise projector, task head) are not included. Observed 2026-10-07.

40.2±2.3comparison only: no current run option verifiedfrom 6 benchmark results across 6 sources

Score within the LMArena text-to-image comparison set only.

$ how it compares

The image models ranked either side of UniWorld-V1, out of 197 with a score.

#modelscorewhereRAMreleased
142CogView4-6B Z.ai41±6comparison only2025-03
143Midjourney V7 Midjourney40±3comparison only2025-04
144UniWorld-V1 PKU-YuanGroup40±2comparison only2025-06
145UniPic2-Metaquery-9B Skywork AI40±6comparison only2025-08
146Show-o2 1.5B-HQ Show Lab, NUS40±3local~7.9GB2025-06

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

$ what it's good at

text to image▁▂▇▇▆▅█▇▂▃▁▁ #84 of 157 * · 11% of its score · 2 tests

Beats Latent Diffusion text2img-large (LDM-KL-8-G, LAION-400M), VQ-Diffusion (ITHQ), Ovis-Image-7B here, though they rank higher overall.

testresultvs the others in its rundifficultycountswhy
WISE (WiScore overall, GPT-4o judge) lab55.0% ✎▄·▆▂▆█▄▆▂▄·▄ beats 17/22medium · avg 60%▇▇▇▇ 11%at its level
GenEval lab80.0% ✎▇▃▅▇▁▄▅▃█▃▃▅ beats 26/39easy · avg 79%—not counted yet: no other model in the same run to compare it with
image to image / editing▁▁▁▃▂▃█▃▃▃▁▁ #75 of 81 · 65% of its score · 5 tests

Trails OmniGen (v1) here, though they rank lower overall.

testresultvs the others in its rundifficultycountswhy
Edit-Compass (EN, overall)34.2% ✎▃▃▂·▃▆▂█·▃·▃ beats 5/20hard · avg 45%▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇ 37%typical
UnicBench (EN, overall)56.0% ✎▂····▃▄▂▄·█▂ beats 1/11easy · avg 73%▇▇▇▇▇▇▇ 18%at its level
CREval-Bench creative image editing (Overall, GPT-4o QA evaluator)65.4% ✎▄█▄··█··▄·▄▄ beats 0/8easy · avg 74%▇▇▇▇ 10%blunt · at its level
GEdit-Bench-EN overall (G_O, GPT-4.1 judge) lab48.5% ✎▄·····█·▄▄·▄ beats 0/5—not counted yet: no other model in the same run to compare it with
ImgEdit-Bench overall (GPT-4.1 judge) lab65.2% ✎no other model yet—not counted yet: no other model in the same run to compare it with
other tests2 not in a task · 22% of its score
testresultvs the others in its rundifficultycountswhy
RealPref-Bench image editing (human user-study score)65.3% ✎█▄▄····▄···▄ beats 0/5easy · avg 78%▇▇▇▇▇▇▇▇ 19%at its level
ViGoR-Bench zero-shot visual reasoning, image models (Result Avg)29.0% ✎▂···▂·▂█▆▅·▂ beats 6/15hard · avg 31%▇ 3.5%typical

$ where to run it

No verified API or download recorded yet · add one

Weights and documentation: huggingface.co/LanguageBind ↗

$ how the score was worked out · Edit-Compass & EditReward-Com… 37%, HP-Edit: A Human-Preference P… 19%, UnicEdit-10M: A Dataset and B… 18% and 3 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
Edit-Compass & EditReward-Compass: A Unified Benchmark for Image Editing and Reward Modeling1▇▇▇▇▇▇▇▇▇▇▇ 37%CC-BY-4.0
HP-Edit: A Human-Preference Post-Training Framework for Image Editing (arXiv:2604.19406v1)1▇▇▇▇▇▇ 19%CC-BY-4.0
UnicEdit-10M: A Dataset and Benchmark Breaking the Scale-Quality Barrier via Unified Verification for Reasoning-Enriched Edits1▇▇▇▇▇ 18%CC-BY-4.0
WISE: A World Knowledge-Informed Semantic Evaluation for Text-to-Image Generation (arXiv v4) · lab-reported1▇▇▇ 11%CC-BY-SA-4.0
CREval: An Automated Interpretable Evaluation for Creative Image Manipulation under Complex Instructions (arXiv:2603.26174v1)1▇▇▇ 10%CC-BY-4.0
ViGoR-Bench: How Far Are Visual Generative Models From Zero-Shot Visual Reasoners? (arXiv:2603.25823v1)1▇ 3%CC-BY-4.0
prior · pull toward the middle▇ 2%
weighted average= 40.2

not counted (3)

PKU-YuanGroup — UniWorld-V1 technical reportGEdit-Bench-EN overall (G_O, GPT-4.1 judge) [correct]48.5%fewer than three models have this test yet, or this source ran it for this model alone
PKU-YuanGroup — UniWorld-V1 technical reportGenEval [correct]80.0%fewer than three models have this test yet, or this source ran it for this model alone
PKU-YuanGroup — UniWorld-V1 technical reportImgEdit-Bench overall (GPT-4.1 judge) [correct]65.2%fewer than three models have this test yet, or this source ran it for this model alone

$ 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