The best AI image generators in 2026, free and paid

Almost every big AI app will now turn a sentence into a picture, and most will do it for free up to a point. They differ in the small print: how much the free tier gives you, whether a watermark goes on the file, who can see your images, and whether you may sell the result. We went through the plans, help pages and terms of ChatGPT, Google Gemini, Microsoft Copilot, Midjourney, Ideogram, Leonardo, FLUX, Adobe Firefly and Canva on 30 September 2026, and picked one for each job.

The nine compared

Prices are the vendors’ US list prices per month unless marked, as shown to us on 30 September 2026. Only Leonardo and Canva put a firm number on their free allowance; the others say “limited” or “may vary” and leave it there. We left out xAI’s Grok this time, which we cover in Grok vs ChatGPT.

Tool	Current image model	Free tier	Paid from	Commercial use
ChatGPTOpenAI	ChatGPT Images 2.5	Limited, number not published	Go $8, Plus $20	Yes, you own output
GeminiGoogle	Nano Banana 2; Nano Banana Pro on paid plans	Daily quota, number not published; 1K downloads	AI Plus $4.99	Google claims no ownership
CopilotMicrosoft	MAI-Image-2 rolling out	Included, number not published	Microsoft 365 Personal $9.99	Personal use only
Midjourneymidjourney.com	V8.2	None on web or Discord	Basic $10 ($8 yearly)	Yes; Pro or Mega above $1M revenue
Ideogramideogram.ai	Ideogram 4.0	Weekly slow credits, amount varies	Plus $20 ($15 yearly)	Yes
Leonardoleonardo.ai	Lucid Origin, Phoenix and partner models	150 tokens a day	Essential $12 ($10 yearly)	Free: licence only
FLUXBlack Forest Labs	FLUX.2 [klein], [pro], [flex], [max]	None confirmed	From $0.014 per image	Yes, via the API
FireflyAdobe	Firefly Image 5 and partner models	Daily generations, number not published	Standard $9.99	Yes, for Firefly models
Canvacanva.com	Canva AI, Standard to Ultra quality	Up to 20 AI uses a month	Pro US$144 a year	Yes, any lawful purpose

The chat assistants all include image generation on their free plans without saying how many images that means. The specialist tools are more precise about quotas and stricter about rights: Midjourney has no free plan on its website, and Leonardo keeps broad rights over what free users make.

Chat assistants that make images

ChatGPT

OpenAI’s image tool is ChatGPT Images 2.5, released on 8 September 2026 with sharper detail, more precise editing and faster generation. It’s available on every tier. You can describe an image, start from a template such as a poster, or sketch something on mobile and say what it should become. Editing works on your own uploads too: select an area and describe the change. ChatGPT can also add text, make the background transparent and work in any aspect ratio.

Free has “limited image creation” and Go at $8 has more. Plus at $20 adds images with thinking, where the model plans and refines before it generates, and Pro, from $100, is listed as “unlimited and faster”, subject to abuse guardrails. OpenAI’s terms assign you ownership of what you create. Every image carries C2PA Content Credentials and an invisible SynthID watermark, with no visible mark unless you ask for one. If you’re choosing a chatbot as well, see our list of ChatGPT alternatives.

Google Gemini

Gemini’s image model is Nano Banana 2, and that’s what the free plan uses too. Google lists better text rendering in all supported languages, character consistency, local edits and combining several uploaded photos into a new one. Without a paid plan, downloads come in 1K. Any Google AI plan raises that to 2K and adds Nano Banana Pro, reached through “Redo with Pro” when an image needs more detail, particularly text. Editing is only open to over-18s.

There’s a daily quota, but Google doesn’t publish the figure. AI Plus costs $4.99 a month for twice the free usage, AI Pro $19.99 for four times, and Ultra $99.99 or $199.99. Gemini images carry a visible watermark as well as SynthID. Google’s terms say it “won’t claim ownership” of content you generate. The same Google AI plans come up in our comparison of which AI assistant is worth paying for.

Microsoft Copilot

Copilot makes images from a prompt or edits an upload you describe. Microsoft said in March that its own MAI-Image-2 was rolling out in Copilot and Bing Image Creator. The free app includes image creation, Microsoft 365 Personal at $9.99 a month gives “higher usage than free”, and Premium at $19.99 gives “extensive usage”, with no numbers attached to either. Read the consumer terms before using the pictures for work. They say “You can only use Copilot for your own personal use”, and forbid removing the content credentials Copilot adds. Fine for a birthday card, then, but not for a client.

Dedicated image generators

Midjourney

Midjourney’s default model is V8.2, released on 24 July 2026 with a focus on aesthetics and personalisation. It brought a new Edit Model that replaces Omni Reference, Character Reference and Retexture, and it keeps V8.1’s HD mode for 2K images. There’s no free plan. The help centre says “No free trial is currently available in Discord or the midjourney.com website”; only the niji·journey app has a limited trial.

Basic costs $10 a month for 3.3 hours of fast GPU time. Standard at $30 adds unlimited Relax mode images, Pro at $60 and Mega at $120 add Stealth Mode, and paying yearly takes 20 percent off. You own what you make, but a company grossing more than $1,000,000 a year needs Pro or Mega to use the images commercially. Without Stealth Mode, your images appear in Midjourney’s public gallery.

Ideogram

Ideogram made its name by getting words right inside images, and version 4.0 is pitched the same way: “Crystal-clear type”, multilingual text, precise layout control and realistic 2K images. It has also been released with open weights on GitHub and Hugging Face, and commercial deployments of the model itself need a separate licence.

Eligible free accounts get weekly slow credits, and Ideogram says the amount may vary. Email-and-password sign-ups get none; use Google, Apple or Microsoft. Free images are public, since private generation starts with Plus at $20 a month ($15 billed yearly), which includes 1,000 priority credits and unlimited slow ones. Pro costs $60 ($42 yearly) with 3,500. Ideogram’s pricing FAQ says you can use your images “for any purpose, including commercial use”.

Leonardo

Leonardo states its free allowance as a plain number, 150 fast tokens a day, reset every 24 hours and never carried over. What a token buys depends on the model and settings, and the generate button shows the cost before you start. Besides its own Lucid and Phoenix models, Leonardo runs third-party ones such as Nano Banana Pro and GPT-Image-1, which always cost tokens.

Essential is $12 a month ($10 billed yearly) for 8,500 tokens, Premium $30 ($24) for 25,000 plus unlimited relaxed generation on Leonardo’s own image models, and Ultimate $60 ($48) for 60,000. Paid subscribers keep full ownership. On the free tier, Leonardo says it “holds rights to use, reproduce, modify, and distribute any images you create”, and you get a non-exclusive, royalty-free licence to use them commercially.

FLUX

FLUX, from Black Forest Labs, runs behind the scenes in other apps; Adobe Firefly offers FLUX.2 [pro] as a partner model. Bought directly, it’s a pay-as-you-go API with a browser playground and no subscription. The FLUX.2 family is priced per image by resolution, from $0.014 for [klein] 4B, through $0.03 for [pro], to $0.07 for [max]. BFL’s newer FLUX 3 does video today and lists images as “soon”.

BFL’s developer terms say you own your outputs and may use them “for your or their own personal or commercial purposes”. The downloadable FLUX.2 [dev] weights are marked non-commercial, and running them in a business requires a paid licence. We couldn’t find a free playground allowance on BFL’s own pages, so treat it as paid.

Design tools with AI built in

Adobe Firefly

Firefly runs Adobe’s own Firefly Image 5 alongside partner models: Gemini 3.1 with Nano Banana 2, GPT Image 1.5 and FLUX.2 [pro]. The free plan offers “limited free daily generations” across standard and premium features without saying how many. Standard costs $9.99 a month with 2,000 generative credits. Pro at $19.99 has 4,000 plus Photoshop on web and mobile and Adobe Express; Pro Plus is $49.99 and Premium $199.99. Every paid plan includes unlimited standard image generation, Generative Fill and Generative Expand; credits go on video, audio and partner models.

Adobe’s commercial-use wording is the clearest of the nine. It says “Outputs from Adobe Firefly AI models are safe for commercial use”, because the models are trained on licensed and public-domain content, and it attaches Content Credentials automatically. That covers Adobe’s own models only; with partner models, Adobe says it’s up to you whether the output is right for your project. Firefly makes video too, and we looked at its free video allowance in our guide to free AI video generators.

Canva

Canva’s advantage is that the image lands in the editor you’ll use to lay out the poster. Canva Free gives you up to 20 uses of Standard or Premium AI tools a month, shared across every AI feature. Pro, at US$144 a year, no longer counts Standard-quality generations against your allowance (fair use applies) and gives up to 200 Premium or 20 Ultra uses. Ultra, the highest image quality, is paid-only. Canva’s AI terms say you own your output and may use it “for any lawful purpose”, unless it builds on content from Canva’s library, in which case that content’s own licence applies.

How we chose

Watermarks, labels and who owns the picture

Most of these tools label their images, even when you can’t see the label. OpenAI adds C2PA Content Credentials and SynthID, Google adds SynthID plus a visible watermark in Gemini, Adobe attaches Content Credentials, and Copilot’s terms forbid stripping its credentials. Assume that a platform can tell your picture was made with AI.

Free plan or paid plan

Verdict

Frequently asked questions

Sources and method