whatmodel.sh / image / InstructPix2Pix

InstructPix2Pix

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

InstructPix2Pix is an open-weights image model from UC Berkeley (Brooks, Holynski, Efros), released January 2023. Its published size is 0.86B parameters. Licence: mit.

Catalogue identity notes

timbrooks/instruct-pix2pix (instruct-pix2pix-00-22000), the only public InstructPix2Pix checkpoint; Stable Diffusion v1.5-based image editor (0.86B UNet as in SD 1.x). HF card licence MIT. Memory estimated as for SD 1.x. Not MagicBrush (a separate fine-tune).

18.8±3.2#143 of 144 image modelsfrom 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 InstructPix2Pix, out of 197 with a score.

#modelscorewhereRAMreleased
192OneCAT-3B Meituan (OneCAT team)24±5local~25GB2025-09
193LCM Dreamshaper v7 SimianLuo (LCM authors)23±4local~1.7GB2023-10
194InstructPix2Pix UC Berkeley (Brooks, Holynski, Efros)19±3local~2.6GB2023-01
195X-LXMERT UNC Chapel Hill17±4comparison only2020-09
196AttnGAN (COCO) Microsoft / Rutgers / Duke16±4comparison only2017-11

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

$ what it's good at

image to image / editing▃▁·▃▂▃█▃▃▄▁▁ #62 of 62 · 81% of its score · 4 tests
testresultvs the others in its rundifficultycountswhy
UnicBench (EN, overall)29.2% ✎▃····▂▄▂▄·█▂ beats 0/11easy · avg 73%▇▇▇▇▇▇▇▇▇▇▇▇▇ 33%at its level
Edit-Compass (EN, overall)23.8% ✎▃▃▂·▃▆▂█·▃·▃ beats 0/20hard · avg 45%▇▇▇▇▇▇▇▇▇ 22%too hard for it
ImagenWorld text-guided editing (human overall)45.0% ✎█··█··██···█ beats 0/4easy · avg 73%▇▇▇▇▇▇ 15%blunt · at its level
MultiTextEdit cross-lingual text-in-image editing (SemAvg, LVM judge)23.8% ✎▄·▄▄▄▄··█▄·▄ beats 0/8medium · avg 50%▇▇▇▇▇ 11%typical
other tests2 not in a task · 15% of its score
testresultvs the others in its rundifficultycountswhy
EdiVal multi-turn editing overall (EdiVal-O after turn 3)13.0% ✎▃·▂▃·▂·▃▂·▃█ beats 0/14medium · avg 42%▇▇▇▇▇ 12%blunt
PaintBench precise editing (mIoU, 20-task average)0.3% ✎▃·····█▅▃·▃▅ beats 0/9very hard · avg 4.3%▇ 2.3%too hard for it

$ where to run it

Other ways to run it

  1. Run it as published · needs ~2.6 GB, NVIDIA only · from huggingface.co/timbrooks
    uv run --with diffusers --with transformers --with accelerate --with torch python -c "import torch; from diffusers import StableDiffusionInstructPix2PixPipeline; from diffusers.utils import load_image; p=StableDiffusionInstructPix2PixPipeline.from_pretrained('timbrooks/instruct-pix2pix', torch_dtype=torch.float16, safety_checker=None).to('cuda'); p('turn him into cyborg', image=load_image('https://raw.githubusercontent.com/timothybrooks/instruct-pix2pix/main/imgs/example.jpg'), num_inference_steps=10, image_guidance_scale=1).images[0].save('out.png')"

Weights and documentation: huggingface.co/timbrooks ↗

$ how the score was worked out · UnicEdit-10M: A Dataset and B… 33%, Edit-Compass & EditReward-Com… 22%, ImagenWorld: Stress-Testing I… 15% 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
UnicEdit-10M: A Dataset and Benchmark Breaking the Scale-Quality Barrier via Unified Verification for Reasoning-Enriched Edits1▇▇▇▇▇▇▇▇▇▇ 33%CC-BY-4.0
Edit-Compass & EditReward-Compass: A Unified Benchmark for Image Editing and Reward Modeling1▇▇▇▇▇▇▇ 22%CC-BY-4.0
ImagenWorld: Stress-Testing Image Generation Models with Explainable Human Evaluation on Open-ended Real-World Tasks1▇▇▇▇▇ 15%CC-BY-4.0
EdiVal-Agent: An Object-Centric Framework for Automated, Fine-Grained Evaluation of Multi-Turn Editing (arXiv:2509.13399v3)1▇▇▇▇ 12%CC-BY-4.0
MultiTextEdit: Benchmarking Cross-Lingual Degradation in Text-in-Image Editing (arXiv:2605.08163v2), Table 31▇▇▇ 11%CC BY 4.0
PaintBench: Deterministic Evaluation of Precise Visual Editing (arXiv:2606.00188v1)1▇ 2%CC-BY-4.0
prior · pull toward the middle▇ 4%
weighted average= 18.8

$ 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