Skip to content
Claudexia TeamREVIEW

AI image generators: 9 services, what is free, licenses

AI image generators: Alisa AI, GigaChat, Shedevrum, ChatGPT, Gemini, Qwen, Midjourney, FLUX, Stable Diffusion. What is free, country notes, who owns the output.

You pick an AI image generator by three things: whether it opens from your country, what is free in it, and whether you may use the result beyond personal purposes. Below are nine services, each with only what its official pages say. Which one draws best, we do not know. We ran no tests of our own, and we do not trust other people's ratings that come without a method.

Nine services: free tier and countries

ServiceFree, per the vendor's pageWhat is said about countries
Alisa AI (Yandex)Image generation is available for free; you can ask for 4 variants and set the aspect ratioA Russian service. Alisa Plus can be connected only in Russia and Belarus, and with a Belarusian IP address Studio (Студия) and some other features of the option may be unavailable
GigaChat (Sber)The help center calls the service free; images are drawn by the Kandinsky modelLogin by phone or Sber ID. For people outside Russia the FAQ names a Russian IP and Sber ID
Shedevrum (Yandex)The app is free, generations go through Yandex ID. The paid Pro subscription opens additional modelsThe help articles we opened have no country list. The service is for people over 18, teenagers only in Kids mode (Детский режим)
ChatGPT (OpenAI)ChatGPT Images is available on all plans, the free plan has a separate limit on generations, the help center gives no numberRussia is not on the ChatGPT and API country list
Gemini (Google)Nano Banana models in the Gemini app, there is a daily image quota, the help center gives no numberRussia is not on the country lists of the web app and the Android app
QwenOn the qwen.ai home page the assistant is called free and open to everyone, with image generation among its featuresThe qwen.ai home page has no country list, and we did not look for one separately
MidjourneyNo free plan: four paid plans from $10 a month, a limited trial exists in the niji · journey appWe did not look for a country list. For payment from Russia, see our separate article on paying for Midjourney from Russia
FLUX.2 [klein] 4BOpen weights, free under the Apache 2.0 license, runs on your own GPUThe model card names no countries
Stable Diffusion 3.5Open weights, free under the Stability AI Community License for research and non-commercial use, and for commercial use at annual revenue under $1 million; the threshold is counted together with affiliatesThe license lists no countries, but it requires compliance with applicable law, including export controls and trade sanctions

The terms of the free tiers change for every service, so check the current numbers in the service itself. Alisa AI and GigaChat have separate caveats, covered below. The help centers of ChatGPT and Gemini give no daily numbers for images. ChatGPT's help center promises a notice when you reach the limit, while Gemini, per Google's help center, warns you already as you approach it.

The OpenAI page with the country list warns that access from outside the list may lead to an account block, and OpenAI publishes no separate list of unsupported countries. We give no legal assessments and do not promise that any method is lawful or safe for your account. You read and accept the vendor's terms. We do not describe ways around blocks or region spoofing.

Rights to the output

ServiceWhat is writtenWhere
Alisa AIContent may be used for personal non-commercial purposes; with an Alisa Plus subscription, texts and images may be used for other purposes tooTerms of 10 September 2026
ShedevrumAll rights to the created objects belong to Yandex. Free use is allowed for non-commercial purposes with a link to the post or a "made in Shedevrum" mark (сделано в Шедевруме), and commercial use requires a request. The Pro subscription allows commercial use without approvalFAQ and the Pro subscription page of the help center
ChatGPTThe user keeps the rights to the input and owns the output to the extent the law allows. Results may not be unique: others will get similar onesOpenAI Terms of Use, in force since 1 January 2026
GeminiThe help center reminds you that you accepted Google's terms and prohibited use policy and must not violate other people's copyright and privacy. This page says nothing about who owns the outputHelp page on image generation
MidjourneyCommon commercial terms on all plans. Under section 4 of the terms, a company or any of its employees with annual revenue strictly above $1 million must be subscribed to Pro or Mega in order to own the images they create. The plans help page states the threshold the same way: more than $1 million in gross revenue a yearSection 4 of the Midjourney terms and the help page "Comparing Midjourney Plans"
FLUX.2 [klein] 4BThe open weights may be used commercially under Apache 2.0. The 9B version comes under a non-commercial licenseModel card on Hugging Face
Stable Diffusion 3.5Research and non-commercial use is free at any revenue. Commercial use is free for individuals and organizations with annual revenue under $1 million (or the equivalent in local currency), and it requires registering with Stability AI. The threshold is counted over you and your affiliates, separately or together, and it does not matter whether the revenue came from Stability AI models. An affiliate, by the agreement's definition, is any entity that directly or indirectly controls you, is controlled by you, or is under common control with you. Control means direct or indirect ownership or management of more than 50% of voting interests. Once revenue exceeds $1 million, the licenses under the agreement terminate, and you must request a license from Stability AI, which may grant it at its discretionStability AI Community License of 5 July 2024
GigaChat, QwenIn the web interface the GigaChat help center does not name rights to images. The GigaChat API has a free mode only for personal non-commercial use. We did not find Qwen's termsSber documentation pages

Rights to the output do not answer the other question, whether your image infringes someone else's rights. The Midjourney help center adds that an upscaled image made by another user belongs to its author.

What they can do and how the limits work

Alisa AI. It draws avatars, icons, prints, postcards and illustrations. It can edit an uploaded photo (JPEG or PNG) and animate it in Studio mode. You cannot upload a picture and ask it to "draw something similar". Generated images stay in the chat history for 14 days, and with Alisa Plus for 30 days. Images edited in Studio lie in the My works section (Мои работы) until you delete them yourself. Without Alisa Plus, image generation can be switched off for some users at peak load, while text stays on. More about the service in our review of Alisa AI.

GigaChat. According to the giga.chat help center (article changed on 29 April 2026), Kandinsky 3.1 sits inside. Sber's blog on Habr (a Russian tech publishing site) from 28 April 2026 says Kandinsky 6.0 Image Pro is available in Telegram, MAX (a Russian messenger), on giga.chat and in the Android app. We did not check which version you have now. Review of the service: GigaChat.

Shedevrum. A free app for Android 7.0 and newer and iOS 15.0 and newer, plus a website. Without signing in you can only browse the feed, and generating requires a Yandex ID. It understands requests in any language and answers in Russian. The help center lists forbidden topics in requests: specific people, politics, religion, violence and adult content. The Pro subscription upgrades images to 4K and lets you download without a logo. The sources disagree on the number of image variants. The subscription help page says six instead of two, while the Yandex Plus option terms that took effect on 17 November 2026 say two. We did not check how many variants the app shows.

ChatGPT. The "Images in ChatGPT" help page names the current version as ChatGPT Images 2.6. It creates images, edits uploaded ones, can make a background transparent, and accepts edits to the whole image or to a selected area. More in our articles on creating photos in ChatGPT and on GPT Image 2.

Gemini. The help page describes three model variants: Nano Banana Lite (for fast, simple images), Nano Banana 2 (a balance of speed and quality, accepts several references) and Nano Banana Pro for reprocessing with a Google AI plan. On a personal account generation is available from age 13, or the applicable age in your country, and editing from 18. On a school account access may be restricted differently. The free options are broken down in our articles on Nano Banana for free and on Nano Banana Pro.

Qwen. The Russian version of the qwen.ai page describes the assistant as free and open to everyone. The open weights of Qwen-Image-2.1 on Hugging Face carry the Qwen Research License Agreement, and we did not go through its terms. More in our Qwen Image 2.1 article.

Midjourney. Plans are Basic $10, Standard $30, Pro $60 and Mega $120 a month (page updated on 3 November 2026). There is no free trial on the website or in Discord. A limited trial exists in the niji · journey app for iOS and Android (article of 10 December 2025). Look for free alternatives in our breakdown of Midjourney alternatives.

FLUX.2 [klein] 4B. Per the model card it fits in about 13 GB of VRAM and runs on an RTX 3090 or 4070 and up. The 9B version and FLUX.2 [dev] come under a non-commercial license.

Stable Diffusion 3.5. The model card on Hugging Face is open, while the weight files become available after you accept the license terms. The card does not name the VRAM amount, so we give no requirement for it. The license terms are described above.

If you also need animation, see our guide to animating a photo with AI and our piece on AI video tools. Another generator, Seedream, has its own article, and a selection for working with finished photos is in AI for photos for free.

How to choose: five questions

QuestionWhat to look atWhat we found on the pages
CountryWhether it is on the service's listOpenAI and Google have lists, and for the other services we found none
LimitWhether a number is namedWe found no daily image count for any service
RightsPersonal or commercial useThe table above: the terms differ a lot
Prompt languageWhether it understands RussianShedevrum accepts requests in any language, Gemini has Russian in the language list of its web app, and we did not look at the others
DataWhere your photos and requests goRead each service's privacy policy. What not to send at all is covered in our article on LLMs

How to write prompts, per vendor recommendations

The Gemini help page advises starting a request with a word like "draw" or "create", describing the style (photorealism, watercolor, cartoon illustration) and listing the character, the action and the background in detail. The Alisa AI help page on text in images advises putting the inscription in quotation marks, writing words the model gets wrong in capital letters and splitting long text into parts. To choose between results, ask for "several variants", and you get four images. A template built from these tips:

Draw a poster for a children's party in a watercolor style.
Characters: balloons and confetti, light background.
Caption: "Fun and GAMES". Format 16:9. Several variants.

If you need API access to image models

Claudexia does not generate images in a chat and does not replace the services above: it is an API gateway. As of 18 November 2026 the Premium catalog has the image models gpt-image-2, gpt-image-2-5, gpt-image-2-6, nano-banana, nano-banana-2 and nano-banana-pro, and Cheap has gpt-image-2, gpt-image-2-5, gpt-image-2-6 and nano-banana-pro. In the live price list the unit for gpt-image-2-6 is tokens: in Premium $3.2 per 1 million input tokens and $4.8 per 1 million output tokens. In Cheap zeros come back (0 / 0), and we do not know what they mean. For the other image models the unit and rates are not given. The price list does not say how many tokens one image takes, so the cost of one image cannot be derived from it. See the models page. How to choose between the two families is covered in the article on GPT Image and Nano Banana as an image API.

You get a key with the sk_cdx_ prefix after registering with a login and password or through Google, GitHub or Telegram. There is no email field. You can pay in rubles through SBP (Russia's Faster Payments System), with a Russian card or in crypto, and the minimum top-up is $1.

For this article we sent no API requests and generated no images.

FAQ

Which AI image generator is free and works without a VPN?

Alisa AI, GigaChat and Shedevrum are Russian. The GigaChat help center begins by asking you to turn off the VPN and sign in with Sber ID, and for Alisa AI and Shedevrum we found nothing about a VPN. ChatGPT and Gemini do not name Russia in their country lists. We give no legal assessments.

Can I sell images made by AI?

It depends on the service, and the terms are in the rights table above. Alisa AI allows personal non-commercial use for free, and Shedevrum allows non-commercial use with a link to the post or a "made in Shedevrum" mark. Read the vendor's terms before you publish.

Which AI draws best?

We have no answer. We ran no tests of our own, and the vendors publish no comparable numbers. Take one prompt and run it in two services.

Can I run an image model on my own computer?

Yes, if you have enough VRAM. By the card, FLUX.2 [klein] 4B needs about 13 GB. The Stable Diffusion 3.5 card is open but names no VRAM amount, and the weight files require you to accept the license.