Free AI for photos: what actually works from Russia without a VPN

«Free AI for photos» is one of the most searched phrases on the Russian internet, and almost every time it hides two completely different jobs. One person wants an image drawn from scratch out of a text description. Another wants the background cut off a product shot, an old photo sharpened, or a stray cable erased from a desk. Those are different tools, and the word «free» means something different in each case.

Here is the picture as of September 2026: what opens from Russia directly, where you need a workaround, and where the free tier hits its ceiling.

First, decide which job you have

Split the task in two. Generation is when there is no source image and you describe the picture in words: «a ceramic mug on a wooden table by the window». Editing is when the photo already exists and needs changing: background out, sharpness up, an object swapped, skin retouched.

The confusion comes from the fact that many services do both, just unevenly. Kandinsky draws well from scratch and edits other people’s photos poorly, while a dedicated background remover draws nothing at all but cuts objects out more cleanly than any generator. Pick for the job, not for the famous name.

What «free» actually means

Four traps that make a free service stop being free right around the time you get used to it.

• The limit and how it is counted. Daily, weekly, or rolling. Rolling is the annoying one: Playground gives ten generations per three-hour window, so you cannot push out forty images in one sitting.
• The commercial licence. Krea and Playground both restrict the free tier to personal use. An image for a client, a marketplace listing, or an ad formally needs a paid plan, even if you already downloaded it.
• The quality ceiling. remove.bg hands out free results at 612 by 612 pixels. Fine for a preview, useless for a product card or print, and the full-size download costs credits.
• Sign-up and payment. A card is not always required, but a foreign account often is. If a service will not take a Russian card at all, «free» hits a wall at the first upgrade. We have a separate guide on paying for AI tools.

Russian services: open immediately, no card

The dullest and most reliable part of the list. No VPN, no foreign account, no card on file. Quality sits below the top international models, but it covers most everyday and content work.

Shedevrum by Yandex

An image, short-video and text generator built on the YandexART model. It works in Russian, so there is no need to translate prompts. The generation itself lives mostly in the mobile app, the website offers less, and the daily caps, if they exist at all, are soft enough that people rarely report hitting them.

Upside: free, no card, more than twenty styles. Downside: fine detail and complex scenes are clearly weaker than Nano Banana or Seedream, and if you plan to sell the output, read the terms once yourself. Rights to generated work are spelled out as a separate clause and reviews interpret it differently.

Kandinsky by Sber

The other Russian generator, reachable through the Fusion Brain site, the Sberbank Online app and GigaChat. It understands Russian natively and needs neither a VPN nor a foreign card. The free allowance is generous: the figure usually quoted is around a hundred generations a day per user, though Sber adjusts it periodically and there is a queue at peak hours.

The current line is Kandinsky 5: an HD image model plus a model for roughly five-second clips. Kandinsky holds composition and artistic styles noticeably better than Shedevrum, but it is just as bad at text inside the frame. Signs and Russian lettering break reliably.

Kandinsky inside GigaChat

You can ask for an image right in the chat: «draw a cat on a windowsill». GigaChat passes the request to Kandinsky and returns the result, and all you need is a SberID. One real limitation: the chat only outputs 1024 by 1024 squares, and the free daily allowance is noticeably smaller than on Fusion Brain itself. Handy for a couple of images in passing, awkward for actual work.

International tools: free exists, with conditions

Gemini and Nano Banana by Google

Currently the best balance of quality and convenience for photo work: the model follows instructions like «make the mug blue, leave everything else alone» precisely, and it is the only mainstream option that reliably writes readable Russian text inside an image. The Gemini app gives the fast version free, at roughly twenty generations a day. Pro is not available on the free tier.

One big problem: opened directly from Russia, Gemini answers «not available in your country», and Google does not accept Russian cards. Two ways around it, a VPN with a foreign account or a Russian reseller running on the official API and billing in roubles. Model and prompt detail sits in our Nano Banana Pro guide, the head-to-head is in Nano Banana vs Seedream, and how to pick a reseller is in our overview of AI aggregators.

Krea

A workbench that puts dozens of image, video and upscaling models behind one interface. Free tier: 100 compute units a day, no card at sign-up. A hundred units is enough to iterate on a still image, nowhere near enough for video or heavy upscaling. The catch that matters: the free licence is not commercial.

Playground

An infinite-canvas editor, useful when you are assembling a banner or a cover rather than making one picture. Free: ten generations per three-hour window and ten downloads a day, no card needed. Across a busy day that adds up to forty or fifty images, but only in batches. Like Krea, the free tier is personal use only.

Ideogram

Known for handling text inside images better than average, which is why people reach for it on posters and logos. There is a free tier, but it is small and runs on «slow» credits that queue behind paying users. Ideogram changes the exact credit count more often than reviews get rewritten, so check the pricing page rather than an article. Russian lettering still comes out mangled.

Midjourney: there is no free tier

Midjourney shut its free trial down back in April 2023 and never brought it back. The cheapest plan is around ten dollars a month and there is no legitimate cheaper route. The one official exception is roughly twenty free generations for new users in the niji mobile app, which is a separate anime-focused product. Everything else advertising «Midjourney for free» is either a shared account or a reseller.

Editing photos: a completely different shelf

If you already have the source image, you barely need generators. You need narrow tools, and those are the ones most likely to be free.

Removing backgrounds

The classic remove.bg works, but the free output is a 612 by 612 preview and the full-size file costs credits. The good news is that background cutout technology went open years ago, and dozens of free services do it just as well. Photoroom and Canva both open from Russia, the only friction is payment. Practical advice: run one typical image of yours through three services and keep whichever handles hair and semi-transparent edges best. That is where the difference actually shows.

Increasing resolution

The best free option here is not an online one at all. Upscayl is an open-source app for Windows, macOS and Linux that runs locally on your GPU: no internet, no sign-up, no limits, no watermark. It upscales up to 16 times and ships several Real-ESRGAN-based models tuned for photos, illustration and anime, with batch processing for a whole folder.

The only requirement is a reasonably capable graphics card. In exchange, your files never leave your machine, which matters on its own if you are handling someone else’s photos. There are plenty of online upscalers too, but almost all cap you at three to five images a day.

Swapping objects and retouching

Free options are thin here. Precise instruction-based editing («remove the cable, leave everything else») is exactly what the Google models are good at, and those do not open from Russia directly. Russian generators handle it more bluntly: they will repaint a chunk of the frame in the general spirit of the original rather than patch it cleanly. For a one-off job, paying a reseller a few dozen roubles per generation beats spending half a day fighting a free tool.

And one job people routinely file under editing: turning a still photo into a short video. That is different models and different money, covered in our separate guide.

Free forever: your own machine

If you are willing to spend one evening on setup, Stable Diffusion running locally removes the limit question entirely: no queues, no external content filters, no ambiguity about rights. Hardware demands have dropped, and a card around the RTX 3060 level covers basic work. The honest downsides: installation takes that evening, out-of-the-box quality sits below the top cloud models, and anything the cloud does with one button you assemble by hand here. A good path if you need images regularly and in volume. A bad one if you need three a month.

The short answer, by task

• An image from a Russian prompt, fast, no setup - Shedevrum or Kandinsky. They open immediately and need no card.
• Readable text inside the image - Google models only, via VPN or a reseller. Russian generators cannot do this yet.
• Background removal - any of the free removers, tested on your own image with hair or glass in it.
• Upscaling - Upscayl, locally. Free, unlimited, and nothing gets uploaded anywhere.
• Lots of images every month - Stable Diffusion on your own machine, or a paid plan. Free limits die on day one at that volume.

A free tier limits the number of attempts far more than it limits quality. And a good image almost never arrives on the first try. It arrives on the fifth.

Frequently asked questions

What is the best free AI for photos with no VPN?

For generating from scratch in Russian, Kandinsky by Sber: a generous allowance, native Russian, nothing to configure. Shedevrum is simpler but weaker on complex scenes. For editing an existing photo the comparison is meaningless, because narrow tools beat generators there.

Can I use free generations commercially?

It depends on the service, and it is worth checking before rather than after. Krea and Playground restrict the free tier to personal use outright. Russian services put rights to the output in a separate clause of their terms, and reviews read that clause differently, so go to the source. For paid client work, a paid plan is primarily a legal purchase rather than a technical one.

Do I need a card to sign up?

Not for Shedevrum, Kandinsky, Krea, Playground or Upscayl. A card only comes up when you upgrade, and that is exactly where a Russian one usually fails.

Why does a free image look so obviously AI-generated?

That is not about the free tier, it is about defaults: models drift toward blown-out gloss, everything in focus, plastic-looking skin. The fix is in the prompt. Add a line describing the optics and film stock at the end: lens type, shallow depth of field, fine grain, lifted blacks, no sharpening, no HDR. That single line changes the result more than switching services would.

The free limit ran out mid-task. Now what?

Three options in ascending cost. Move to another free service and finish there, if the style allows it. Pay a reseller per generation, in roubles, without a subscription. Or buy a monthly plan if you need images regularly, which starts paying off somewhere around a few hundred images a month.

Is it worth learning this properly?

If you need images and video for work rather than as a one-off, yes. The difference in results comes far less from which service you pick than from how the request is phrased and how the process is set up. We run training on AI video and graphics covering exactly that: prompts, access from Russia, and assembling the finished material.