The poster was already up before anyone told me. Not online, either, a real printed thing, taped to a noticeboard, and it said GRAND OPNEING in big cheerful letters, and I stood there reading it twice because my brain kept autocorrecting it to what it was supposed to say. A friend eventually messaged me, very politely, and I took it down, and that was the afternoon I stopped trusting any image tool with words in it. Which is a strange place to start a guide to image tools, I know, but honestly that poster explains more about this whole topic than any feature list, because every one of these tools will hand you something beautiful with one stupid flaw hiding in it, and the flaw is always the part you didn’t check.
Anyway. The poster came from Ideogram, or rather, it came from me not understanding Ideogram yet. See, most image tools are terrible with text, genuinely terrible, they will spell things with total confidence and get them wrong, and Ideogram is the one exception I have found, the one that can actually manage a few readable words, a shop sign, a poster headline, a book cover title. Two or three words, that’s the sweet spot. I went back to it later for a book cover, just the title, nothing else, and it got it right on the second try, which after the poster felt like a small miracle. But a full sentence? A slogan with punctuation? No. It melts into alphabet soup like everything else, and I don’t care how good a tool claims to be, if you ask it for a paragraph inside a picture you are asking for trouble. Read every letter. Every single one. Even when you’re sure.
That poster night was not my first disaster, though. The first one was quieter. I needed one image, one, for a blog post I had already finished writing, told myself five minutes, and three hours later I was still at my desk with cold tea and a cat that appeared to have a hand growing out of its shoulder. The cat came from Microsoft’s Image Creator, inside Designer and Copilot, which is the tool I open first now, most days, because it is the least fussy thing I have used. Sign in with a Microsoft account, type, and it gives you several images at once, and that matters more than it sounds, because choosing the best of four is a completely different feeling from staring at one weird result and trying to talk yourself into it being fine.
But that first night I typed “a cat” and got the most average cat imaginable, and then I typed slightly different versions of “a cat” about twenty times, because I hadn’t yet learned the only lesson that actually matters with prompts, which is that the tool doesn’t understand your sentence, it guesses, and it fills in every gap you leave with its own choices, and its choices are not yours. So you have to give it the details yourself. Not fancy language, just: what is it, exactly, a ginger cat, not a cat. Where is it, a sunny windowsill, not nowhere, because no setting gets you a cat floating in beige nothingness, and I have generated that, more than once, I wish I hadn’t. What style, a real photo look, an illustration, watercolour, pick one, because if you don’t pick, it picks. And what kind of light and mood, soft morning light, calm, because that changes the whole feeling of the same scene, and that was the last piece I figured out, months in, which tells you something about how I learn things.
Oh, and the cat with the hand. Faces and hands. That’s the other thing nobody warns you about properly. Hands have so many possible positions that the pattern just goes blurry, and eyes end up looking in two directions at once, and at thumbnail size you will miss all of it, your eye slides right past. My proof is a thumbnail I put up one Tuesday without a second look. Small, it was perfect. A reader opened it properly that evening and messaged me about the hand — six fingers, plain as day, on a character I had been chatting with Gemini about for the post itself. That chat is the appeal, really: you are already mid-conversation, you ask for a picture, you ask for a tweak, you never leave the tab, and brainstorming a look for something feels effortless. Effortless is the trap, though. But the thumbnail lied. So now, anything that is going to be published gets opened at full size first, and I look at it properly, hands, faces, background corners where melted objects hide, the whole thing, about thirty seconds per image, which is nothing compared to the embarrassment it saves. The background, by the way, is where the weird stuff lives. You check the face, you check the hands, and then there’s an extra limb behind a chair that you only notice after publishing. Ask me how I know.
Where was I. Right, Canva. The trouble with Canva is that it is already open. If your slides and your social posts live there, then of course the generator lives there too, tucked into the same sidebar, and walking away from that convenience feels ridiculous. Picture, drag, your own text on top, export — no downloads in between, no colour shift between apps, no friction at all. Friction, it turns out, was doing me a favour. The catch, and there is always a catch, is that the free plan only gives you a limited number of generations, and I burned through mine in a single afternoon making test images of a reading nook I never used, not once, not for anything. So now I save them for pictures I actually need, decorative illustrations, backgrounds, simple scenes, and here’s the trick I wish I’d known sooner: for anything with words on it, generate just the background with AI and add the text yourself in Canva, so the spelling and the font and the placement are yours, and nobody has to squint at a poster that says WELCMOE. Which, yes, also happened. Different poster. I don’t want to talk about it.
And then there’s the one I only open when money is involved, or a client might see the thing, or I just don’t want to lie awake wondering where an image was trained. Adobe Firefly. Adobe built it around licensed content, and they actually talk about commercial use in plain language instead of mumbling about it in terms and conditions, which makes it the safer choice for a website or a client project or anything that earns money, and the free tier is a monthly allowance, so I save up ideas and do them in one batch, list on paper next to me, very organised, very unlike me the rest of the time. The pictures are clean and professional and a bit formal, product-style scenes, tidy illustrations, less playful than the others, which is a plus for a business topic and a minus when I wanted something wild and got something fit for a corporate brochure. If rights matter to you, and they should once money is involved, read Firefly’s current terms before you publish, and keep a note of which tool made which image. I keep mine in the download folder now, after a month where I could not remember whether a header image came from Firefly or somewhere sketchier, and that month taught me more about record-keeping than any productivity article ever did.
Oh, and one more thing about Firefly, I almost forgot, which is sort of the point of this whole guide: the terms change. Free plans change. A tool that was generous in January can quietly tighten up by June, and if you build a deadline around a free tier without checking, you will spend the morning of publication scrambling, which I have done, and I do not recommend it. Check the official site, not a blog post from last year. Including, if I’m honest, this one, eventually.
Leonardo. Right. Leonardo is the one my more artistic friends kept recommending, and I finally opened it, saw about forty settings staring back at me, and closed the tab again. That was day one. The free tier runs on daily credits, there are more style presets than I knew what to do with, fantasy art, character designs, game-style illustrations, detailed scenes with actual atmosphere, and all of that is wonderful once you learn it, and completely overwhelming until you do. What finally worked was embarrassingly simple: one preset style, one simple prompt, change one setting at a time, stop when it looks good. For a story, a game project, a portfolio — anything where the art itself is the point — I would put up with the learning curve now. I would not have said that on day one, when I closed the tab. If you need one illustration for a post going out tomorrow morning, open something simpler and save yourself the tutorial spiral. I say this as someone who fell into the tutorial spiral.
So, if I had to sum it up without a spreadsheet, because I tried the spreadsheet version and it read like a spreadsheet: Microsoft’s is the generous, easy one for quick blog illustrations, and its weakness is text and faces going strange. Canva’s lives inside your design workflow, limited generations, save them for real needs, shines for social graphics. Firefly is the careful one, monthly allowance, less playful, more formal, the one I trust most when rights matter. Ideogram is the text specialist, daily limits, keep the words to a few and it will surprise you. Leonardo is the detailed-art one, daily credits, wonderful once you learn it, confusing until you do. Gemini is the quick-ideas one inside a chat, free tier with limits, and you must check faces and hands at full size. Knowing which weakness you can live with matters more than which feature list is longest, and I mean that, I picked tools by their weaknesses in the end, not their strengths, because every strength here is roughly the same and the weaknesses are what actually bite you.
There are also places I simply do not use AI images, and I’m firmer about this now than I was at the start. Not as a real product photo, if you’re selling something, because the buyer deserves to see the actual thing they’re buying, not a prettier imaginary version of it. Not as a photo of a real person. Not as evidence that something happened, because that isn’t illustration any more, that’s fabrication with better lighting. Not to copy a living artist’s distinctive style and pass it off as your own, and not a trademarked character or logo in a commercial project without understanding the rights, because “the AI made it” is not a defence anyone will accept. For all of those, use a real photo, a licensed stock image, your own design, something true. I keep saying “illustration, not evidence” to myself ever since I saw a seller use an AI picture for a handmade item and watched the comments turn — people felt tricked, and honestly they were right to.
A Few Questions I Keep Getting Asked, Answered Unevenly
Are free AI image generators really free? Short answer, yes, but with strings attached that only show up when you need the tool most. You get a daily or monthly cap, or smaller downloads, or fewer styles, and the terms quietly move — I once planned a whole week’s posts around a tool that had tightened its free tier while I wasn’t looking, and spent a Monday morning redoing every header image. Read the official free plan page first. Not a review. The page itself.
Can I put AI images on my blog or my channel? Usually, yes — with a “check first” attached. Commercial use is where tools split: Firefly is built for it, others are vaguer, and the answer can differ image by image. Boring advice, I know, but the alternative is a takedown notice after the post is already shared, and I would rather spend five minutes reading terms than a week undoing that.
Why do AI images get hands and text wrong? Think about what the tool is actually doing: it is copying patterns from pictures, not learning bones or an alphabet. A hand can sit a hundred ways, so the pattern smears — six fingers, a thumb in the wrong place, eyes drifting apart. Letters are worse, because one wrong stroke ruins the whole word, and the model has no idea what “wrong” means. Skies are easy. A person holding a sign is not, which is why your landscape looks flawless and your poster looks like something from a strange dream.
Should I say when an image is AI-made? Platforms increasingly want you to, and readers respect it even when they don’t ask. My rule now is simple: if someone could look at it and think it is a real photo, it gets a label. After the posters, that is self-preservation as much as honesty.
Conclusion, or Whatever This Turned Into
So which one should you actually open? It depends on the job, a frustrating answer but the honest one, and the one my first night taught me. Quick blog illustrations, start with Microsoft’s Image Creator or Canva. Commercial-style work where rights matter, look at Firefly and read the terms. Text inside the image, try Ideogram and keep the words short, and for heaven’s sake read every letter before you publish. Detailed art, give Leonardo a real afternoon, not ten minutes. And if you are already chatting with Gemini and want a quick idea, fine — just do not trust the thumbnail. Open it properly first.
These tools will not turn you into a designer in a weekend. They did not do that for me, and I have spent months on them. What they will do is let you dress up an idea quickly and cheaply, which, for a blog post going out tomorrow, is often enough. So: made-up pictures for made-up scenes, real photos for real things, and one honest look at the full-size result before anyone else sees it. And if you ever print a poster, maybe read it twice. Trust me on that one.
Leave a Reply