Free of charge and free to use are two different questions, and the second one decides whether you can use the finished image at all. Most of the big providers have some kind of free access, but each of them limits something different. Google delivers images from the Gemini apps at 1K instead of 2K without a subscription, and keeps a visible watermark in the picture on the free and Pro tiers, the Gemini sparkle. Microsoft Designer is free to use, but according to Microsoft's own help page it is "licensed for personal, non-commercial use". Adobe offers free daily generations for Firefly without a plan and writes about its own models: "Outputs from Adobe Firefly AI models are safe for commercial use." Midjourney currently has no trial in Discord or on the website, and anyone working at a company with more than one million US dollars in annual revenue needs a Pro or Mega plan under Midjourney's terms in order to own the images they generate. On top of that sits the third question, which no provider answers for you: whether a purely generated image is covered by copyright at all. The German Federal Ministry of Justice writes in its FAQ from March 2024 that purely AI-based content does not enjoy copyright protection. All details in this article: as of September 2026.
I do not generate AI images as an end in themselves, I generate them as supply: as B-roll in a reel, as the background of a graphic. So they do not end up in a folder here, they end up in an edit, and the generation itself is a cost item I approve before a larger series runs. Everything I collect about AI at work is on the AI page; how the roles are split up in my business is in My AI Workforce, and how I decide what gets handed off at all is in Which Tasks You Can Hand Off to AI.
"Free" means one of three things
Before you compare tools by series consistency, text in the image and German, it pays to separate the three kinds of free. They feel identical on the first image and clearly different by the twentieth.
- A permanent free tier with a cap. You can keep generating, but the result is trimmed: smaller resolution, a visible watermark, a daily limit. That is the case with the Gemini apps.
- An allowance that gets used up. Credits or boosts that refill monthly or daily. Once the allowance is empty, things get slower or stop until the period resets. That is the model at Adobe Firefly and at Microsoft.
- Open model weights on your own hardware. No fee to a provider, but compute time, electricity and setup on your side. FLUX.1 [schnell] from Black Forest Labs is released under Apache 2.0 and is therefore open to commercial use as well. Its sibling FLUX.1 [dev] comes with a licence for non-commercial applications.
The third kind comes closest to "no sign-up": once the model has been downloaded, generation can run locally without a service account. The download itself still depends on the platform and its access rules. Hosted services from the large providers usually require an account; age limits are a separate condition, with Google's help page stating a minimum of 13 years for image generation in the Gemini apps and 18 years for image editing.
What the free tiers limit
Three observations. Resolution is the least conspicuous limit and, in case of doubt, the most expensive one: a 1K image is usually fine for a reel, for print or a large website hero it gets tight. A watermark is not just cosmetics, and Google says it keeps the visible sparkle for exactly that reason, to make the images easier to identify as AI-generated. And an allowance changes how you plan: at ten attempts per motif, a daily limit arrives faster than it looks on the first prompt.
What the terms say about commercial use
This is where providers differ more sharply than they do on features, and it is rarely the part printed on the landing page.
Microsoft Designer is the clearest case. The official FAQ says: "Designer is licensed for personal, non-commercial use with your Microsoft account." That is a statement about Designer, not about every Microsoft product with an image feature. Anyone planning to illustrate client work with it should check the applicable terms Microsoft links from that page first.
Adobe Firefly goes the other way and advertises precisely this. Adobe writes that outputs from the Firefly models are "safe for commercial use" and backs that up with training material from licensed and public domain content. That is a vendor statement, not a court ruling. Beta features, uploaded references, prompts and the concrete output can still raise separate contractual or rights questions.
Midjourney ties ownership to the plan. The terms say you own the assets you create "to the fullest extent possible under applicable law", subject to the obligations of the agreement and the rights of third parties. For companies above one million US dollars in annual revenue, a Pro or Mega plan is the precondition for that.
Open models depend entirely on their licence. FLUX.1 [schnell] is under Apache 2.0 and therefore open commercially, FLUX.1 [dev] is not. Stability AI's Community License permits commercial use as long as annual revenue stays under one million US dollars; above that, Stability requires an enterprise licence.
Google uses its help page on image generation mainly to cover resolution, age limits and the reminder not to violate anyone else's copyright or privacy rights, and otherwise points to the terms of service. So anyone using Gemini images commercially reads those directly instead of relying on a summary.
Labelling: what is inside a generated image
Providers now mark their outputs with methods you cannot see. Google states that all media generated with Google's tools carry an imperceptible SynthID watermark; the visible sparkle on the free and Pro tiers comes on top of that and, according to Google's announcement of 20 November 2025, is removed for Ultra subscriptions and inside the AI Studio developer tool. Microsoft applies Content Credentials based on the C2PA standard to images created with Designer's editing features, which can be read at contentcredentials.org. OpenAI describes in its help article that images from ChatGPT receive C2PA metadata, that this metadata can be stripped by platforms, editing tools or file conversions, and that a SynthID watermark is embedded as a complement for that reason.
On the platform side, Meta labels content with an "AI info" note when it detects industry-standard markers or when someone discloses AI use while uploading, and says it does not add the note for minor changes such as cropping or colour correction.
Legally, Article 50 of the EU AI Act has applied since 2 August 2026. The duty to mark synthetic images in a machine-readable format is addressed to the providers of the AI systems, not to you as the user. For systems that were already on the market before that date, the European Commission names a transition period until 2 December 2026. A disclosure duty of your own applies mainly to deepfakes, and according to the Commission the marking duty does not apply where the system only performs an assistive function in standard editing.
Who owns an AI image
The German Federal Ministry of Justice answers the question in its FAQ "Künstliche Intelligenz und Urheberrecht" from March 2024 like this: works protected under German copyright law are only the personal intellectual creations of a human being, so purely AI-based content does not enjoy copyright protection. Authorship on the user's side would come into consideration "at most" where the software is merely a tool and plays a subordinate role in the creation process. And further: it has to be assessed case by case whether sufficient influence over the concrete form remains in human hands. The document is from March 2024 and therefore older than the date of this article; the ministry itself writes in it that it is pushing for an open review of whether AI-generated output needs an adapted legal framework. So where the line sits in an individual case is open, and this article is not legal advice.
In practice that means two things. First, if a purely generated image has no copyright protection of its own, you generally cannot derive an exclusive copyright from that fact. Second, the absence of protection is not permission to use the image freely. Service terms, trademarks, personality rights, confidentiality and rights in recognisable source works can still restrict use. The ministry points out that when using an AI output you have to check whether pre-existing works are still recognisable in it, and that personality rights can be violated when an image creates the impression of being an authentic depiction of a person. A generated logo, a recognisable third-party motif or a face that belongs to someone remain sensitive, whichever service produced the picture.
Where an image like that ends up in my business
A single AI image is not a result here, it is a building block. The editing runs through Eddi, my AI employee for video: he receives raw footage and a script and returns a finished, technically checked video, with subtitles, overlays and in 9:16. He works on the command line with ffmpeg, a locally running Whisper model and Remotion, he does not shoot anything himself, and he never overwrites raw footage. For AI image generation the same rule applies as for publishing: it does not run without my approval, because it costs money. How he works and where his limit sits is on his page, Eddi. Which program handles the edit around it is something I went through in AI Video Editing Software Compared.
The distinction that matters is the same one as with video: a generated image is material, a finished reel is a work result, and in between sits the work nobody fits into a prompt.
Frequently asked questions
Can you create AI images without signing up?
Without a service account, what remains is mainly the local route via open model weights, for example FLUX.1 [schnell] under Apache 2.0. Once downloaded, it can run locally; the download itself may still require access through a platform. Hosted services from the large providers usually require an account. Google separately states minimum ages of 13 for image generation in the Gemini apps and 18 for image editing. A local model costs you setup and compute instead.
Can I use free AI images commercially?
That depends on the service, not on the price. Microsoft explicitly writes "personal, non-commercial use" for Designer. Adobe states that outputs from the Firefly models are safe for commercial use. Midjourney ties ownership to the plan and, above one million US dollars in revenue, to Pro or Mega. Check the terms of the service you are using before the image goes onto a client page.
Are there AI images without restrictions?
Not with the hosted services. Google says on its help page that it may remove images when its systems detect a possible policy violation, and the other providers have usage policies too. A locally running model takes the provider's check out of the loop, not the legal responsibility: copyright, trademark and personality rights apply regardless of where the image was made.
Which app works on a phone?
Google's Gemini app and Microsoft's Copilot app generate images directly on the phone, Adobe Firefly runs in the browser and in Adobe's own apps. For a reel, the resolution of the free tiers is often enough. If you plan to display the image large later, check what your tier delivers before you download.
How can you tell an image came from an AI?
Through provenance signals the providers embed, and they only work in one direction. If you find a marker, you have a signal: Google points to SynthID for its outputs, and Content Credentials under C2PA can be read at contentcredentials.org. If you find none, that does not prove human origin because metadata can be lost during uploading, editing or conversion. A metadata check also says nothing by itself about the licence or the image content.
How to take it further
Take an image you need this week anyway and run the test once from end to end: generate, download, check the resolution, check the metadata, and only then decide whether it goes onto a page you earn money with. Those two minutes on the licence question are the only place where a free image gets expensive later.
If you want to see how an image like that travels into a finished video instead of sitting in a downloads folder: that exact sequence, from raw footage through the edit to approval, is what I show in my community, along with the finished templates for the employees who handle it in my business.
