Shedevrum: Yandex's Free AI for Images and Short Video

You know the scene. It's evening, you sit down to publish a post, the text is done, and there's no picture. None of your own photos fit the topic, you've seen the stock shots of smiling people in suits a hundred times, and your designer will answer tomorrow at best. So the post goes out with the first image you can find, or it doesn't go out at all.

AI image generators close that gap in a couple of minutes. Shedevrum is one of the biggest of them that most Americans have never heard of: Yandex's app for pictures and short videos, with millions of installs and a public feed where you can peek at the exact prompts behind other people's images. It's free to start, and for a post cover, a story background, an article illustration, or a client greeting card it's more than enough.

I'll be upfront about one thing. Shedevrum is a Russian service, built for a Russian audience, with Russian payment rails. Some of it won't apply to you, and I'll flag every spot where that's the case. The good news: about two-thirds of this guide – the prompt formula, the 15 prompts, lettering tricks, video tips, troubleshooting, and the pre-publish checklist – works the same in any image generator you use. Read in order, or jump to the section you need from the table of contents.

Section 01What Shedevrum is in 2026

Shedevrum is Yandex's service for generating images, videos, and texts. The name comes from the Russian word for "masterpiece." You describe what you want to see in words, for example "a ginger cat asleep on a windowsill, a winter city outside, soft morning light," and a few seconds later you get finished versions of the picture. Images and video come from Yandex's YandexART family of models, and texts come from YandexGPT.

If Yandex is new to you: it's Russia's largest tech company, roughly what Google is in the US – search, maps, email, taxis, music, and a voice assistant called Alice. Shedevrum is its consumer AI art app.

The service launched in April 2023. Since then the app has gathered millions of installs, and by the end of 2023 it ranked among the three most popular AI apps in the world. Inside, Shedevrum works like a social network: every user has a profile, you can publish your work to a shared feed, like it, follow authors, and see the prompts other people used.

What matters as of early October 2026:

• The models got a serious upgrade. YandexART 2.5 came out in March 2025, version 2.7 in summer 2025, and on July 14, 2026 Yandex released Alice AI ART 2.0. This model both generates images and edits them from an instruction, and it also powers images inside Yandex's Alice AI chat. In Yandex's own internal comparison the model took second place in the world for text-to-image generation. That's the company's claim, not an independent benchmark, but the jump from versions two years old is visible to the naked eye.
• Text on images finally works. After the May 2026 update, according to Yandex, the model writes Russian words correctly three times more often. Yandex's numbers are about Russian lettering, so for English text, test it yourself. There's a separate section below with techniques that help in any generator.
• There's video too. Shedevrum makes clips a few seconds long: it draws the first frame, then animates it. You can choose between several video models in the menu, and some are subscription-only.
• There's a paid tier called Shedevrum Pro. It costs about 100 rubles (roughly a dollar) a month on top of Yandex Plus and removes the watermark, skips the queue, adds 4K, and allows commercial use. Details in the next section.

Where Shedevrum lives: the website shedevrum.ai and apps for iPhone and Android in the App Store and Google Play. The service is meant for users 18 and older; there's a separate kids' mode with coloring pages, and you can only exit it with a password.

The interface and most of the community are Russian-speaking, so expect the feed to be full of Russian prompts. That's actually useful: paste any prompt into a translator and you get a free lesson in how other people describe light, style, and framing.

Section 02What it costs: free vs Pro

The core of Shedevrum is free. You sign in with a Yandex ID and start generating right away. The free tier has these limits: each request returns 2 versions of the image, at peak hours you wait in a queue, there's a daily generation limit, and downloaded images carry a Shedevrum watermark in the corner.

Yandex's help pages don't currently state the exact daily limit, and it has changed over time: when the subscription launched in 2025, the site listed 70 free generations, with a noticeably lower limit for video. For normal work, a few covers for your posts, the free limit is plenty. The people who hit it are the ones cycling through dozens of versions in a row.

Look at the last row; it matters more than all the others. Under Yandex's rules, images from the free tier may be used for personal, non-commercial purposes. A cover for a post that sells your consultation, an illustration on your course landing page, an ad creative – that's commercial use. For those you need Pro. I cover rights in detail in Section 12.

How to turn on Pro, if it's available to you: in the app, open your profile and tap the Shedevrum Pro banner; on the website, the subscribe button sits in the profile menu. For Russian users this is one of the service's big conveniences – rubles, a local card, no foreign payment hassle. For everyone else it's the main friction point.

Section 03How to start: site, app, and Yandex ID

There's no separate Shedevrum registration. You sign in with a Yandex ID, the same account Russians use for Yandex Mail, Music, or Taxi. If you don't have one, Yandex creates it from a phone number in about a minute.

Does Shedevrum work from the US? Honest answer: Yandex has no official statement about access from outside Russia. Inside Russia the service opens directly. Outside, it depends on your country, your app store, and whether Yandex's sign-in accepts your phone number. I'm not going to suggest workarounds that break a service's terms. If it opens for you, great; if it doesn't, skip to Section 14 and pick one of the alternatives – everything you learn about prompts in this guide carries over.

Section 04Your first image in 3 minutes

The most common beginner mistake goes like this: someone types "coffee" or "woman in an office," gets something bland, and decides the AI is weak. The model draws exactly what you described, and the word "coffee" describes very little. Here's how to get a picture you'd actually put in a post.

Look at what worked here. Subject and place set the scene. Details turn an abstract "coffee" into a specific morning. Light and tones create the mood. The words "close-up photo, shallow depth of field" tell the model how to hold the camera. Strip it all back to "coffee" and you get an image that fits every post in the world – and therefore none of them.

One more useful habit: save prompts that work to your notes. In a month you'll have your own small library, and covers in a consistent style will take a minute. If you want ready wording for different tasks, browse my AI prompt library and the collection of AI photo prompt ideas.

Section 05The prompt formula and 15 ready prompts

A prompt is the text request you give an AI. For Shedevrum, and honestly for any image generator, a simple six-part formula works. You don't have to fill in all six every time, but the more parts are in place, the more precise the picture.

Shedevrum understands requests in any language, so you can write in English. Its strongest language is Russian, which makes sense given its audience, but plain, concrete English descriptions work. One to three sentences is the sweet spot. Too short and you get an average picture; too long and the model gets confused and drops some of the details.

Below are 15 prompts you can copy and adapt. I built them around the tasks of people who sell services, courses, and consulting: covers, backgrounds, illustrations, greeting cards. They're written to be generator-agnostic, so they'll work just as well in ChatGPT, Gemini, Midjourney, or Ideogram.

Notice there are no people with faces in these prompts. That's deliberate. AI draws people well, but faces and hands still come out strange sometimes, and Shedevrum doesn't allow images of real, well-known people at all. Objects, landscapes, and metaphors come out reliably on the first try, and for a post they work just as well as a portrait.

Section 06Styles and formats for each platform

Shedevrum has preset styles: photo, anime, pixel art, oil painting, watercolor, comic, Soviet-era poster, and others. You can pick a style with a button in the menu or write it out in the prompt. Writing it out gives you more control: "watercolor with soft bleeds on white paper" gives a more precise result than the plain "watercolor" button.

Choose the aspect ratio before generating; you can't change it afterward without making a new image. Shedevrum offers square 1:1, vertical 3:4 and 9:16, and horizontal 4:3 and 16:9. Here's how that maps to platforms:

A trick for a post series: keep one shared "tail" on every prompt. For example, end every cover prompt for the month with "minimalism, pastel lavender and mint tones, soft diffused light, top-down view." Only the subject at the start changes, and your feed looks like a designer runs it. I go deeper into building covers in one style in my piece on the AI banner maker.

There's also a separate mode for your own photos. You upload a picture and Shedevrum redraws it in one of eight styles: plush toy, knitted figure, cartoon, pixels, and others. For a personal brand it's a fun move: a cartoon-style avatar for a holiday post, or a "knitted" version of your team for an introduction post.

Section 07Text on images without typos

For a long time every AI image generator wrote gibberish on pictures. With the 2026 update, Yandex's model got noticeably better at lettering. Mistakes still happen, but with the right prompt a short caption comes out right on the first or second try.

Yandex itself published three tips for lettering, and they work in other generators too:

My working rule: give short captions of 1-3 words to the AI, and put long headlines with numbers, dates, and prices on top of the image yourself in Canva, an Instagram story editor, or any design app. The AI draws the background, you add the exact text. That way no date in an announcement drifts and no price quietly turns into a different number.

Section 08Video: bringing a picture to life

Shedevrum makes short clips a few seconds long. It works in two steps: first the model draws a starting frame from your description, then it turns that frame into a moving video. The finished clip saves as an MP4, and you can download it after publishing.

What this kind of video is good for: a moving story background, a short intro before a live stream, an animated reels cover, an animated greeting card for a client. You won't get a full clip with a storyline and speech out of Shedevrum – that takes other tools – but it does a few seconds of beautiful motion well.

Simple, slow motion works best: steam, water, clouds, leaves, fire, a smooth camera push. Fast, complex actions like a running person or dancing still come out shaky. Keep one motion per clip and the result stays clean.

If you need longer video with voice or editing, see my breakdown of AI video from text in minutes. Different tools, different approach.

Section 09Your own photos, Studio, and edits

With Alice AI ART 2.0, Shedevrum learned to edit images from a plain sentence. You take a finished picture and write what to change: "make the background lighter," "remove the cup on the right," "turn fall into winter." The model edits that part and tries to keep the rest. That saves a lot of generations: before, changing one detail meant redrawing the whole thing and hoping everything else matched.

Pro subscribers get Shedevrum Studio. It's a separate working mode with tools you need when you make images in volume:

• Batch generation. One request gives you many versions at once, and you pick the best ones for a post series.
• Collages. Studio combines several images into one layout, handy for a carousel or a roundup cover.
• Portrait retouching. You can adjust your own photo, for example the light or the background.
• Background removal. Cuts out an object or a person so you can place it on another background or a banner.
• Your own seed. A seed is a number that fixes the randomness of a generation. With the same seed and a similar prompt, images come out similar, which makes a consistent series style easier to hold.
• Five built-in models. You can run one task through different models and pick whichever did best.

Pro also lets you animate your own photo: upload a picture and Shedevrum turns it into a short clip with gentle motion. For an introduction post or a holiday greeting it feels warm and unexpected.

A separate warning about your own photos. Upload only images you have the rights to, and only people who agreed to it. A client's photo used without asking, even "just to make it pretty," violates their rights, and Shedevrum's own rules forbid depicting real, existing people in the feed.

Section 10Let another AI write your prompts

If writing descriptions feels like a chore or you run out of words, hand it to another AI. I use Claude for this: it holds a style well, gets the task from half a sentence, and produces a whole batch of prompts for a post series at once. You describe the task in your own words and copy the finished prompts into the image generator.

Then you paste the prompts into Shedevrum one at a time and get a series of covers in one style. If one image came out wrong, go back to Claude and write: "the second one is too dark, rewrite it lighter." After a couple of rounds you'll have your own tested style "tail," and you can ask Claude to remember it for the weeks ahead.

How to get Claude: sign up at claude.ai with a Google account or email. The free plan is more than enough for writing image prompts. For more on how to phrase requests, see my guide to Claude prompts: a formula and ready examples.

ChatGPT or Gemini will do the same job if that's what you already use. The skill that matters is the brief: what the posts are about, what to leave out, and one shared style line. Any good chat model turns that into usable prompts.

Section 1110 ways these images work for your business

An image in the feed has one job: stop the scroll so the person reads your text. Here's where an AI image generator saves the most time for people who sell services, courses, and consulting.

Here's the catch: the last two rows and almost every work image in this table count as commercial use. In Shedevrum that means Pro. If Pro isn't available to you, make your business visuals in a generator with a commercial license you can pay for. For the full landscape of tools, see the best AI tools of 2026.

Run the numbers on your own channel. Say you publish four posts a week and spend half an hour hunting for an image for each one – two hours a week. With an AI generator it looks different: on Sunday you ask Claude for four prompts for the week, and in 15 minutes you have covers in one style. Give the hour and a half you free up to replying to followers in your DMs, because that's where client requests are born.

Section 12Usage rights: what Yandex allows

Please read this section carefully. Shedevrum's rules are stricter than most users think, and plenty of people put free images into ads without knowing they're breaking the agreement.

Here's what Shedevrum's user agreement (version dated April 10, 2026) and Yandex's help pages say:

• Free tier – personal, non-commercial use only. Images, videos, and texts from the free tier can be used for yourself. The help pages add that non-commercial use requires a link to Shedevrum or to the post with the work.
• Commercial use without a subscription requires Yandex's written consent. According to the help pages, rights to generated content belong to Yandex, and commercial use needs a separate request.
• On Pro, commercial use is allowed without approval. The subscription page says so directly. For work tasks it's the simplest and most honest route.
• Yandex gets a license to your content. By publishing work, you give Yandex a free, non-exclusive license to use it. If a prompt or image is part of your trade secrets, Pro lets you hide it from the feed.

What counts as commercial in practice? Anything that helps you sell: a cover on a post that sells a consultation, a course banner, a landing page image, an ad creative, the cover of a paid guide. A personal card for your mom or an avatar on a private account is non-commercial.

What you can't generate at all, even on Pro: drugs, weapons, violence and cruelty, gambling, erotica, politics and religion, shock content, and specific real people. Moderation runs on three levels: an automatic filter, a moderator, and a review panel, and complaints are reviewed within 24 hours. Requests on banned topics simply won't run.

Section 13When something goes wrong

Here are the most common problems and what to do about them. Most of these fixes apply to every image generator.

Section 14Shedevrum vs Midjourney, ChatGPT, Gemini, and others

Shedevrum isn't the only option, and different tasks suit different services. For a US reader the real question is simple: is Shedevrum worth adding next to the tools you can already pay for? Here's how it sits next to the popular ones in October 2026. I'm sticking to what's widely known about each and leaving out prices, since they change often – check the current plan pages before you subscribe.

How to choose. You need an image fast inside a chat you already use – ChatGPT or Gemini. You need clean text on a cover – try Ideogram first. You need the most artistic image for a course cover and you're ready to pay – Midjourney. You want the cleanest story on commercial licensing – Adobe Firefly. More on Gemini in my guide on how to use Gemini, and the free options are collected in free AI tools.

Where does Shedevrum fit, then? It's a free playground with a very generous feed of prompts to learn from, a quick way to make short moving backgrounds, and a solid option if you work with Russian-speaking audiences. For a US business, I'd treat it as a second tool for experiments, and make the images you'll actually sell with in a service whose license you can pay for.

Personally, I'd keep two tools in my bookmarks: one image generator for covers and backgrounds, and Claude for the texts and the prompts that feed it. That pair covers nearly all the visuals for a channel, at a couple of hours a week.

Section 15Pre-publish checklist

Run through this list before you put an image in a post, an ad, or on your site:

• The format matches the platform: 1:1 for a post, 9:16 for stories and reels, 16:9 for an article cover.
• The image reads on a small phone screen; the main subject is visible without zooming.
• Any text on it has been read letter by letter, with zero typos.
• Dates, prices, and important numbers were added on top in an editor; the AI didn't draw them.
• No extra fingers, strange faces, or objects melting into each other.
• For business use, the image was made on Pro or in a service with a commercial license; for personal use, it links back to Shedevrum.
• No recognizable people without their consent, and no banned topics.
• The style matches the other posts in the series.
• The prompt that worked is saved to your notes for next time.

Section 16The whole path in a minute

1. Open shedevrum.ai or find the official Shedevrum app by Yandex.
2. Sign in with a Yandex ID; if it won't open for you, use one of the alternatives from Section 14.
3. Write the prompt using the formula: subject, setting, details, light, style, camera.
4. Pick the platform's format before generating.
5. Generate and pick the best version; when you miss, change one detail.
6. Keep lettering short, in quotes and capitals; add long text in an editor.
7. For video, describe one slow motion and check the first frame.
8. Ask Claude to write prompts for the whole post series.
9. For business images, use Pro – about 100 rubles (roughly a dollar) a month on top of Yandex Plus – or a generator with a commercial license.
10. Before publishing, run the checklist from Section 15.