whatmodel.sh / image / Kandinsky 2.1

Kandinsky 2.1

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

Kandinsky 2.1 is an open-weights image model from AI Forever, released February 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

· Parameter count ↗

1.22B = the latent diffusion U-Net only. Other stated components: text encoder 560M, diffusion image prior 1B, CLIP ViT-L/14 image encoder 427M, MoVQ 67M (whole pipeline OUR SUM = 3.27B).

No comparable score yet. Its published results are below, but they don't overlap enough with other models to calculate a score.

$ how it compares

Scored image models released closest to Kandinsky 2.1, out of 197 with a score, then more from AI Forever.

#modelscorewhereRAMreleased
166IF-XL DeepFloyd35±2local~29GB2023-04
171IF-L DeepFloyd33±3local~14GB2023-04
177IF-M DeepFloyd31±3local~14GB2023-04
178Dreamlike Photoreal 2.0 Dreamlike.art30±3local~2.1GB2023-01
194InstructPix2Pix UC Berkeley (Brooks, Holynski, Efros)19±3local~2.6GB2023-01
more from AI Forever
107Kandinsky 3 AI Forever48±11local~32GB2023-11
157Kandinsky 2.2 AI Forever37±5local~12GB2023-07

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

$ what it's good at

text to imagenot ranked on this task yet · 1 test
testresultvs the others in its rundifficultycountswhy
COCO-30k zero-shot FID ↓ (Kandinsky 2.1 protocol) lab8.2100 FID ✎█··········█ beats 1/1—not counted yet: no other model in the same run to compare it with

$ where to run it

Other ways to run it

  1. 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-2-1', 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 ↗

$ published results · 1 result, not enough overlap for a score
testresultsource and conditions
COCO-30k zero-shot FID ↓ (Kandinsky 2.1 protocol)8.2100 FIDKandinsky 2.1 card — original COCO-30k FID · lab-reported

Original card Evaluation table, own Kandinsky 2.1 row, FID (30k)=8.21. Zero-shot COCO-30k; generation/sampling details unspecified. Raw distance retained. Fixed latent unit 10 FID, sign reversed because lower is better; single-model source cannot calibrate. Date is review date, not generation date.

$ 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