AI Forever: Kandinsky 2.2

ai-forever/kandinsky-2.2

Access Kandinsky 2.2 from AI Forever using the Puter.js AI API.

Add to your app

Model Card

Kandinsky 2.2 is a multilingual text-to-image model from Sber AI's AI Forever research group, an upgrade over Kandinsky 2 in the same Kandinsky family.

It keeps the two-stage design of a diffusion image prior feeding a latent diffusion U-Net and MoVQ decoder, but swaps in a larger CLIP ViT-bigG-14 image encoder (1.8B parameters) in place of Kandinsky 2's CLIP ViT-L/14 (427M parameters). The stronger encoder improves how closely generated images follow the prompt and raises overall image quality. The update also added ControlNet support for guided generation and increased supported resolution to 1024px with multiple aspect ratios.

It is trained on the LAION HighRes dataset and fine-tuned on roughly 2 million high-resolution, high-quality image-caption pairs for sharper photorealism.

It suits use cases that need higher-resolution output or ControlNet-style guided generation, at the same per-image cost as Kandinsky 2.

Cost Per Image $0.0014

per generation

Configuration second

resolution

Release Date N/A

Add Kandinsky 2.2 to your app

Use Kandinsky 2.2 in your app with the Puter.js AI API, no API keys or setup required.

// npm install @heyputer/puter.js
import { puter } from '@heyputer/puter.js';

puter.ai.txt2img("A serene mountain landscape at sunset", {
    model: "ai-forever/kandinsky-2.2"
}).then(image => {
    document.body.appendChild(image);
});

<html>
<body>
    <script src="https://js.puter.com/v2/"></script>
    <script>
        puter.ai.txt2img("A serene mountain landscape at sunset", {
            model: "ai-forever/kandinsky-2.2"
        }).then(image => {
            document.body.appendChild(image);
        });
    </script>
</body>
</html>

More AI Models From AI Forever

Find other AI Forever models →

Image

Kandinsky 2

Kandinsky 2 is a multilingual text-to-image model built by Sber AI's AI Forever research group, part of the Kandinsky family that also includes Kandinsky 2.2 and Kandinsky 3.0. Instead of the single U-Net Stable Diffusion uses, it generates images in two stages. A diffusion-based image prior maps a CLIP text embedding to a CLIP image embedding, then a 1.22B-parameter latent diffusion U-Net and a MoVQ decoder turn that embedding into pixels. Text is encoded with XLM-Roberta-Large-Vit-L-14, so prompts work across multiple languages, and the image prior uses a CLIP ViT-L/14 encoder. This prior-plus-decoder approach follows the same lineage as OpenAI's DALL-E 2. On the COCO-30K benchmark, this architecture reported an FID of 8.21, ahead of Stable Diffusion 2.1 (8.59), GigaGAN at 512x512 (9.09), and DALL-E 2 (10.39). It fits multilingual prompting and image mixing, and it is a lower-cost, earlier-generation option compared to Kandinsky 2.2's higher-resolution output.

Frequently Asked Questions

How do I use Kandinsky 2.2?

You can access Kandinsky 2.2 by AI Forever through Puter.js AI API. Include the library in your web app or Node.js project and start making calls with just a few lines of JavaScript — no backend and no configuration required.

Is the Kandinsky 2.2 API free for developers?

Kandinsky 2.2 is free to integrate using the Puter.js AI API. With the User-Pays Model, you can add AI to your app for $0, since users cover their own AI usage through their Puter account.

What is the pricing for Kandinsky 2.2?

Kandinsky 2.2 costs $0.0014 per image.

	Price
Per image	$0.0014

Does it work with React / Vue / Vanilla JS / Node / etc.?

Yes — the Kandinsky 2.2 API works with any JavaScript framework, Node.js, or plain HTML through Puter.js. Just include the library and start building. See the documentation for more details.

Add Kandinsky 2.2 to your app for free

Developers can integrate Kandinsky 2.2 for free using the Puter.js AI API.
With the User-Pays Model, each user covers their own AI usage instead of the developer.

Get started How pricing works