whatmodel.sh / image / Kandinsky 2.2
Kandinsky 2.2
↗ share
+ contribute
Kandinsky 2.2 is an open-weights image model from AI Forever, released July 2023. Its published size is 1.22B parameters. Licence: apache-2.0.
Catalogue identity notes
Full text-to-image pipeline including prior/decoder dependencies. No scores copied from third-party model comparisons. Memory is a conservative planning estimate including pipeline components and activation headroom; no local runtime measurement.
Model information sources
1.22B = the latent diffusion U-Net (decoder) only. Other stated components: text encoder 560M, diffusion image prior 1B, CLIP ViT-bigG image encoder 1.8B, MoVQ 67M (whole pipeline OUR SUM = 4.65B).
Score within the LMArena text-to-image comparison set only.
$ how it compares
The image models ranked either side of Kandinsky 2.2, out of 197 with a score, then more from AI Forever.
| # | model | score | where | ||
|---|---|---|---|---|---|
| 155 |  Step1X-Edit v1.0 | 37±4 | comparison only | ||
| 156 |  SDXL-Lightning (4-step UNet) | 37±3 | local | ||
| 157 | Kandinsky 2.2 | 37±5 | local | ||
| 158 |  Janus-Pro-1B | 36±4 | api $0.0005/image +local | ||
| 159 |  Show-o (show-o-demo-512) | 36±3 | local | ||
| more from AI Forever | |||||
| 107 | Kandinsky 3 | 48±11 | local | ||
$ what it's good at
text to image▂▅▅▃▄▄█▃▁▂▁▁ #99 of 126 * · 91% of its score · 2 tests
Beats Latent Diffusion text2img-large (LDM-KL-8-G, LAION-400M), VQ-Diffusion (ITHQ), Stable Diffusion XL 0.9 (base) here, though they rank higher overall. Trails PixArt-α (PixArt-XL-2-1024-MS) here, though they rank lower overall.
| test | result | vs the others in its run | counts |
|---|---|---|---|
| Human text alignment · wins against Kandinsky 3.0 lab | 32.0% ✎ | █·█········█ beats 0/2 | 46% |
| Human visual quality · wins against Kandinsky 3.0 lab | 48.0% ✎ | █····█·····█ beats 1/2 | 45% |
$ where to run it
Other ways to run it
- Run it as published · needs ~12 GB, NVIDIA only · from huggingface.co/kandinsky-community
uv run --with torch --with diffusers --with transformers --with accelerate --with safetensors python -c "import torch; from diffusers import AutoPipelineForText2Image; p=AutoPipelineForText2Image.from_pretrained('kandinsky-community/kandinsky-2-2-decoder', torch_dtype=torch.float16).to('cuda'); p('A lighthouse at sunset', height=512, width=512).images[0].save('image.png')"
Weights and documentation: huggingface.co/kandinsky-community ↗
$ how the score was worked out · Kandinsky 3.0 — original huma… 91%
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
| source | results | share of the score |
|---|---|---|
 Kandinsky 3.0 — original human comparisons · lab-reported | 2 | 91% |
| prior · pull toward the middle | 9% | |
| weighted average | = 36.7 |
$ improve this page
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