Kandinsky and "Alisa draws"

Alongside Shedevrum, the Russian-speaking market has two more notable entry points into image generation: the Kandinsky model and the "Alisa draws" mode. Both are regional, both are free and work in Russian, and both are often a person's first experience with AI pictures, because they're built into services people already use. If you'd rather have one tool instead of several services, you can generate pictures right in the Twelver chat.

Kandinsky

Kandinsky is a neural network from Sber, one of the best-known Russian models. Its strengths:

• Native language and local context — it understands requests without translation and handles recognizable local imagery well.
• Accessibility — there's free access, with no foreign-payment problem.
• An artistic lean — historically Kandinsky is good at illustrative, fantastical, "art" scenes; it's named after the painter for a reason.

Its weak spots are the same as most regional models': extreme photorealism for commercial work trails Midjourney and Flux, and text in the image and complex multi-object scenes come harder.

"Alisa draws"

"Alisa draws" is image generation right inside Yandex's voice assistant. Technically the tech under the hood is related to Shedevrum, but the entry point is different: you just ask Alisa to draw something by voice or text. The query "alisa draws" is popular precisely because millions of people already have Alisa at hand — on a phone, a speaker, a browser — and try generation without installing anything new.

The same prompt you tried in Kandinsky or with Alisa — here, next to the results of other models. Compare each one's "hand".

When to take regional models

A simple rule for choosing between regional and global:

• Take regional ones (Shedevrum, Kandinsky, Alisa) when being free, native-language and barrier-free matters, and the task is everyday or illustrative.
• Look at the global ones (Midjourney, Flux, DALL·E) when you need extreme photorealism or fine control and you're happy to pay.

The practical takeaway

Free regional models probably cover 80% of everyday requests — and do it for free. It makes sense to start with them and move to the global ones only where you've hit the ceiling. It's handier still when both are available in one place: then choosing a model is just a switch for the task, not five separate sign-ups and subscriptions.

What's next

For those who want maximum control and aren't afraid of technical detail, there's a whole separate world — Stable Diffusion and Flux. That's the next chapter.

In the Twelver chat, generation works in one conversation — the same convenience, but with no barriers around access and payment.