Straight answer: pick the AI image generator based on what happens if someone sues you, not on how pretty the sample images look on the homepage. Check the licence for indemnification, test the tool with your own brand prompts before you pay for a year up front, and work out the real cost per usable image at your actual volume, because the headline price and the real price are rarely the same number.
Looking for a free option? See what the Perchance AI image generator can do.
Why this decision matters more than the picture itself
Every business owner I talk to right now wants the same thing: fast, cheap images for ads, product pages, social posts, blog headers, without hiring a photographer or a designer for every single asset. Fair enough. I do it myself for client work. But most people choose their AI image generator the way they’d choose a filter on Instagram, whichever one made the nicest picture in a five-minute trial, then they build a whole content calendar on top of it.
That’s backwards. The picture is the easy part now. Every major tool makes decent pictures. The hard part, the part that costs money when you get it wrong, is the licence sitting underneath the picture.
The candle company and the ghost watermark
I’ll give you a real one. A home fragrance client of mine, a small UK brand doing maybe £40,000 a month through Shopify, needed twelve hero images for a Christmas paid ad campaign. Tight deadline, tighter budget. We generated them in Midjourney because the moodboard results looked gorgeous, warm, cinematic, exactly the vibe she wanted.
One image, when we zoomed in at full size to prep it for print-quality Facebook creative, had a faint ghosted rectangle in the bottom right corner. Blurred, half-formed, but shaped exactly like a stock photo watermark. It wasn’t unique to us either, once you know to look, you start seeing it in other Midjourney outputs online. It’s a strong hint that some training images the model learned from still had watermarks baked into them.
Nobody could tell us with certainty whether that made the final image legally risky. That uncertainty was the problem. We pulled it from the ad set, kept Midjourney for early concept boards and mood exploration, and moved the final paid-ad creative to Adobe Firefly, which is trained only on Adobe Stock, public domain, and licensed content. Slower to generate, less painterly, but I could tell the client with a straight face that Adobe stands behind the commercial licence with actual indemnification. That’s the trade you’re making with every tool on the market, speed and style versus legal certainty, and almost nobody explains it that plainly before you’ve already paid for twelve months of a subscription.
What “commercial use allowed” in the terms covers
Nearly every AI image tool’s marketing page says some version of “commercial use included.” That phrase means far less than people assume. It usually just means the vendor won’t stop you from selling something made with their tool. It says nothing about whether the underlying image accidentally resembles a photographer’s copyrighted work, a brand’s trademark, or a real person’s likeness, because the model was trained on billions of images scraped from the internet, most without permission.
This is the part most buying guides skip because it’s uncomfortable and slightly boring: a “commercial licence” from the tool protects you from the tool’s company. It does not automatically protect you from a third party claiming the output infringes their work. Getty Images sued Stability AI in the UK High Court over exactly this, arguing millions of its licensed photos were used to train Stable Diffusion without consent. A group of illustrators and artists (Andersen v Stability AI, in the US) filed a similar class action against Midjourney, Stability AI, and DeviantArt. Neither case has closed the door on this risk, and none of it has been fully settled in a way that gives small businesses a clean answer.
What protects you is indemnification, a specific clause where the vendor agrees to cover your legal costs if a generated image lands you in a copyright dispute. As of now, that’s offered clearly by Adobe (Firefly), Getty Images’ own AI generator, and Shutterstock’s AI tool, usually only on paid business tiers. Midjourney and OpenAI’s image tools do not offer the same level of cover to individual subscribers. If your images are going on a homepage, a product box, or a national ad campaign, that difference is worth building into your decision, not an afterthought you Google after something goes wrong.
Six things to check before you hand over your card details
- Indemnification, not just “commercial use.” Search the terms page for the word “indemnify” or “indemnification.” If it’s not there, assume you carry the legal risk alone.
- Where the training data comes from. Adobe Firefly and Getty’s generator are trained on licensed or owned libraries. Midjourney and most open models scrape broadly from the public internet. That single fact drives most of the legal risk difference between tools.
- Resolution ceiling for your actual use case. Print work at even a modest 8×10 inch flyer needs roughly 2400×3000 pixels at 300dpi. Midjourney’s base output upscales to around 1024 to 2048px depending on settings; Firefly caps similarly. Fine for a website or social post, often not enough for large-format print without an upscaler.
- Consistency across a set. Generate the same subject ten times with a locked seed or reference image. If your brand character, product mockup, or mascot shifts shape, colour, or proportions each time, that tool will slow you down on any campaign needing more than one matching asset, which is most of them.
- Cost per finished, usable image, at your volume. Not the monthly headline price, the real number after you throw away the six out of ten generations that don’t work.
- How it handles logos, real brand names, and real people. Most tools will refuse or badly mangle requests involving recognisable public figures or trademarked logos, which matters if your prompts ever reference a celebrity endorsement style or a competitor’s product for comparison content.
The step-by-step process I now run with every client
This is the exact sequence I walk clients through before they commit to a subscription:
- Step 1. List every single place the image will end up, paid ads, packaging, print catalogue, YouTube thumbnail, blog header, because licences and resolution needs differ by output.
- Step 2. Pull up the vendor’s terms page and search for “indemnify.” Screenshot it. This becomes the deciding factor for anything customer-facing or paid media.
- Step 3. Run the same ten prompts through your two or three shortlisted tools, using your real brand language, not the demo prompts from the tool’s own marketing.
- Step 4. Count how many of those ten outputs are usable without heavy editing. That’s your real success rate, and it’s rarely above 40 to 60 percent even with a good tool.
- Step 5. Divide the monthly subscription cost by usable images per month at that success rate. A £30 plan that gives you 15 usable images a month is a worse deal than a £10 plan giving you 8, if you only need 6.
- Step 6. Check trademark and likeness risk on anything touching a real person, a competitor’s logo, or a well-known brand aesthetic, and pull those images before publishing if you’re not sure.
The main tools, compared honestly
Prices shift often, so treat these as a snapshot rather than gospel, but the relative positioning tends to hold.
- Midjourney: from around $10 a month for limited fast generations up to $120 for the Mega plan with heavy usage. Best-looking, most painterly output. No indemnification for individual subscribers. Best for mood boards, concept art, early exploration, not the final paid-ad asset if you’re risk-averse.
- Adobe Firefly: around $9.99 a month standalone, or bundled into Creative Cloud plans many businesses already pay for. Trained on licensed content, comes with indemnification on business tiers. Less flashy output, but the one I’d default to for anything going to print or paid media.
- OpenAI’s image tools (via ChatGPT Plus, roughly $20 a month): strong at following detailed instructions and generating text within images, useful for quick concepts and social graphics. No broad indemnification for standard subscribers.
- Ideogram: paid plans from around $8 a month, good at rendering legible text inside an image, useful for packaging mockups and logo concepts where other tools produce garbled letters.
- Leonardo.ai: from around $10 a month, popular with product and game designers, strong consistency tools for repeated characters or products.
- Canva Magic Media: included in Canva Pro at roughly £10 to £13 a month, the easiest option for a non-designer already living inside Canva, but generally the weakest raw image quality of the group.
- Getty Images AI generator and Shutterstock’s AI tool: priced per image, often £15 to £100 depending on plan and resolution, aimed squarely at larger brands that can’t risk any legal exposure and want full indemnification baked in from the start.
If budget is the main constraint before you’re even ready to commit to a paid plan, it’s worth trying the free tiers first. I’ve written up a full list of free AI tools that don’t need a subscription, several of which include enough image generation to test your prompts and workflow before you spend a penny.
What businesses want from these tools right now
When I pulled together our AI Search Demand Report from 3.4 million search queries, one thing stood out on the image generation side: the volume of searches asking specifically about licensing and commercial rights has grown faster than searches for “best AI image generator” itself. People aren’t just looking for a tool anymore, they’re looking for permission to use it without getting a legal letter six months later. That shift tells you the market has already caught up to the risk, even if a lot of the blog content out there hasn’t.
It’s also worth knowing where these images tend to fall apart visually, because it affects your quality checks. I covered this in a separate piece on why AI generated videos sometimes look like dreams, and the same underlying cause shows up in still images: hands with the wrong number of fingers, text that almost reads but doesn’t, reflections that don’t match their source object. Run every commercial image through a hands-and-text check before it goes anywhere public. It takes ten seconds and it’s the single most common mistake I see brands make in a rush.
Practical uses beyond ads and hero images
Commercial use of AI image tools isn’t only about product photography and Facebook creative. I’ve had a client generate a full set of illustrated puzzle backgrounds for a customer loyalty campaign using the same approach outlined in this guide to creating printable word search puzzles for customers, and the image licensing question applied there just as much as it did to paid ads. Print-on-demand products, downloadable lead magnets, packaging mockups, competition entry graphics for something like a promotional giveaway on Related guides you might find useful
