whatmodel.sh / image / Kandinsky 2.0

Kandinsky 2.0

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

Kandinsky 2.0 is an open-weights image model from AI Forever, released November 2022. Its published size is 1.2B parameters. Licence: apache-2.0.

Catalogue identity notes

Comparison-only, proposed so the Kandinsky 2.1 card's COCO-30k FID table counts. 'Kandinsky 2.0 (2022)' in that table = the AI Forever (Sber AI) multilingual latent diffusion model released as ai-forever/Kandinsky_2.0 (sha 04e8fc7ea945965edfcb33705ac272d14b9d0f6c, created 2022-11-22, apache-2.0; card: 'Kandinsky 2.0 — the first multilingual text2image model', 'UNet size: 1.2B parameters', text encoders mCLIP-XLMR 560M + mT5-encoder-small 146M). The 2.1 card says 2.1 kept 'the UNet diffusion block from Kandinsky 2.0'. params_b is the UNet only. Not Kandinsky 2.1/2.2 or the earlier ruDALL-E 'Kandinsky 1.0'.

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 2.0, out of 197 with a score, then more from AI Forever.

#modelscorewhereRAMreleased
110VQ-Diffusion (ITHQ) Microsoft48±15local~4.1GB2022-11
180Stable Diffusion v2.1 Stability AI29±2comparison only2022-12
181Stable Diffusion v2.0 Stability AI29±3comparison only2022-11
182Versatile Diffusion v1.0 SHI Labs29±4local~7.2GB2022-11
185Openjourney PromptHero28±2local~2.1GB2022-11
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) lab20.0000 FID ✎█··········█ beats 0/1—not counted yet: no other model in the same run to compare it with

$ where to run it

No verified API or download recorded yet · add one

Official weights and documentation: huggingface.co/ai-forever ↗

$ published results · 1 result, not enough overlap for a score
testresultsource and conditions
COCO-30k zero-shot FID ↓ (Kandinsky 2.1 protocol)20.0000 FIDKandinsky 2.1 card — original COCO-30k FID · lab-reported

Kandinsky 2.1 card @ 04b3666, '### Evaluation' table 'FID metric values for generative models on COCO_30k', row 'Kandinsky 2.0 (2022)', column 'FID (30k)' = 20.00 (raw distance, lower is better). Same table as the existing Kandinsky 2.1 row (8.21). Kandinsky 2.0 is the same Sber AI / AI Forever team's previous model and the row is not marked as copied, so it is taken as a lab measurement under the brief's same-family rule; caveat: the card's setup sentence ('We quantitatively measure the performance of Kandinsky 2.1 on the COCO_30k dataset, in zero-shot mode') names only 2.1, so the 2.0 figure's generation settings are not restated. Checkpoint: ai-forever/Kandinsky_2.0 @ 04e8fc7ea945965edfcb33705ac272d14b9d0f6c (proposed comparison-only kandinsky-2-0). Date is review date, as for the existing row. Re-read 2026-10-11 at the pinned card (row 'Kandinsky 2.0 (2022)' = 20.00).

$ 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