whatmodel.sh / image / Playground v2.5

Playground v2.5

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

Playground v2.5 is an open-weights image model from Playground, released February 2024. Its published size is 2.57B parameters. Licence: playground-v2dot5-community.

Model information sources

Release date ↗ · Weights licence ↗ · Parameter count ↗

UNet only: HF Hub safetensors total for the official repo (pinned revision), 2,567,463,684 = unet/diffusion_pytorch_model.safetensors (SDXL-architecture UNet). The two CLIP text encoders and the VAE are not included. Observed 2026-10-07.

40.8±1.7#105 of 144 image modelsfrom 7 benchmark results across 7 sources

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

$ how it compares

The image models ranked either side of Playground v2.5, out of 197 with a score.

#modelscorewhereRAMreleased
138Show-o2 7B Show Lab, NUS42±3local~23GB2025-06
139Playground v2 (1024px aesthetic) Playground41±4local~4.9GB2023-12
140Playground v2.5 Playground41±2local~8.8GB2024-02
141Stable Diffusion 3.5 Medium Stability AI41±3api $0.035/image +local~20GB2024-10
142CogView4-6B Z.ai41±6comparison only2025-03

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

$ what it's good at

text to image▂▅▅▃▄▄█▃▁▂▁▁ #85 of 126 · 99% of its score · 7 tests

Beats Latent Diffusion text2img-large (LDM-KL-8-G, LAION-400M), VQ-Diffusion (ITHQ), FLUX.1 Kontext [dev] here, though they rank higher overall.

testresultvs the others in its rundifficultycountswhy
T2I-ReasonBench (overall)50.5% ✎▅▃█▂▂▂····▂▂ beats 3/13easy · avg 69%▇▇▇▇▇▇▇▇▇▇ 26%at its level
GenAI-Arena text-to-image (Elo, 4 Jun 2024)1150.0 ✎▃▃··▃·▃·█·▃▃ beats 8/8avg 1237▇▇▇▇▇▇▇▇▇ 22%sharp
EvalMuse-40K model ranking (FGA-BLIP2 overall alignment)3.20 / 5 ✎█▄·▄██▄▄█▄▄▄ beats 10/14easy · avg 3.83 / 5▇▇▇▇▇▇▇▇ 21%at its level
ConceptMix (k=2 full-mark score)46.0% ✎▄·▄·█▄▄▄···▄ beats 6/7easy · avg 66%▇▇▇▇▇▇▇ 18%at its level
WISE (WiScore overall, GPT-4o judge)49.0% ✎▄·▆▂▆█▄▆▂▄·▄ beats 13/22medium · avg 60%▇▇ 6.1%at its level
KolorsPrompts human overall satisfaction (Kolors team)3.37 / 5 ✎▄·▄········█ beats 1/3easy · avg 3.87 / 5▇▇ 5.3%at its level
HPSv3 image generation benchmark (Overall HPSv3 score)10.2700 ✎▃·····▃··▃·█ beats 3/5avg 10.058▇ 1.0%sharp · noisy

$ where to run it

Other ways to run it

  1. Run it as published · needs ~8.8 GB, NVIDIA only · from huggingface.co/playgroundai (official)
    uv run --with torch --with 'diffusers>=0.27.0' --with transformers --with accelerate --with safetensors python -c "import torch; from diffusers import DiffusionPipeline; p=DiffusionPipeline.from_pretrained('playgroundai/playground-v2.5-1024px-aesthetic', torch_dtype=torch.float16, variant='fp16').to('cuda'); p(prompt='Astronaut in a jungle, cold color palette, muted colors, detailed, 8k', num_inference_steps=50, guidance_scale=3).images[0].save('out.png')"

Official weights and documentation: huggingface.co/playgroundai ↗

$ how the score was worked out · T2I-ReasonBench: Benchmarking… 26%, GenAI Arena: An Open Evaluati… 22%, EvalMuse-40K: A Reliable and… 21% and 4 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
T2I-ReasonBench: Benchmarking Reasoning-Informed Text-to-Image Generation (Findings of ACL 2026)1▇▇▇▇▇▇▇▇ 26%CC-BY-4.0
GenAI Arena: An Open Evaluation Platform for Generative Models (arXiv v3)1▇▇▇▇▇▇▇ 22%CC-BY-4.0
EvalMuse-40K: A Reliable and Fine-Grained Benchmark with Comprehensive Human Annotations for Text-to-Image Generation Model Evaluation (arXiv v2)1▇▇▇▇▇▇ 21%CC-BY-4.0
ConceptMix: A Compositional Image Generation Benchmark with Controllable Difficulty (arXiv v1)1▇▇▇▇▇ 18%CC-BY-4.0
WISE: A World Knowledge-Informed Semantic Evaluation for Text-to-Image Generation (arXiv v4)1▇▇ 6%CC-BY-SA-4.0
Kwai-Kolors/Kolors repository README, human assessment1▇▇ 5%Apache-2.0
HPSv3 repository README, image generation benchmark table (MizzenAI/HPSv3)1▇ 1%MIT
prior · pull toward the middle▇ 1%
weighted average= 40.8

$ 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