You need a header image for a blog post, a clean product shot for an Etsy listing, or something that isn’t a stock photo for a client slide. There’s no budget for a designer and no budget for a stock subscription either. A free AI image generator can cover all three, but only if you stop treating it like a search bar.
The gap between a throwaway free AI image and one you’d actually publish is almost never the tool. It’s the prompt, a couple of settings, and the two minutes of cleanup you do afterwards. Here’s the exact process, with the prompts I use, in the order I run them.
Step 1: Pick the generator to match the job
Different free tools are good at different things. Signing up for five of them and using none properly wastes an afternoon, so start with the one that fits what you’re making.
- Photorealistic people and objects: Bing Image Creator, Leonardo’s free tier, Playground, Ideogram
- Illustration, posters, stylised art: Dream by Wombo, Stable Horde, Craiyon for rough concepts
- Images with readable words in them: Ideogram’s daily free credits, or generate an empty area and set the type yourself afterwards
- All-in-one design work: Canva’s free Magic Media, which also handles templates, resizing and export
Credit limits shift constantly, so it helps to compare current free tiers side by side. This round-up of free AI image generators worth your time covers what each one actually gives you before it starts asking for a card.
Step 2: Write a photo brief, not a search query
Type “dog” and you get a generic dog. Type “greyhound asleep on a linen sofa, warm afternoon light through a window, 50mm lens, shallow depth of field” and you get something you’d happily put at the top of an article.
The five-part prompt formula
- Subject and detail: who or what, plus one specific trait (“weathered leather notebook”, not “notebook”)
- Action or pose: sleeping, mid-stride, holding a mug, pouring water
- Setting: kitchen counter, foggy street at dawn, plain grey backdrop
- Lighting: golden hour, soft window light, overcast, hard flash
- Style and lens: editorial photography, 50mm, flat vector illustration, watercolour
Here’s a complete one for a post about cold brew: “Tall glass of cold brew coffee with ice cubes on a dark oak table, condensation on the glass, morning light from the left, shallow depth of field, 50mm editorial food photography.” Nine out of ten disappointing results come from leaving one of those five parts out. If you want the longer version of this thinking, this breakdown of an AI image creator workflow from blank prompt to finished image goes deeper on structure.
Then fill in the negative prompt
Most free tools give you an “avoid” or “negative prompt” box. Use it every time: text, watermark, extra fingers, deformed hands, blurry, low resolution, oversaturated, logo. It’s the cheapest quality upgrade available.
Step 3: Change one thing at a time
Run the same prompt three times and you’ll get three different images. That’s not a bug, it’s a slot machine. When a result is close but not right, don’t rewrite the whole description. Change one clause and rerun.
Some real edits that work:
- Too dark? Add “bright overcast daylight, no harsh shadows”.
- Background too busy? Swap it for “plain warm grey studio backdrop”.
- Too cartoonish? Add “photorealistic, 35mm film grain”.
- Subject too small? Add “close-up, subject fills the frame”.
Set your aspect ratio before you generate, because cropping later ruins composition. Go 16:9 for blog headers, thumbnails and slides; 1:1 for Instagram feeds and Etsy listings; 9:16 for Reels, Stories and TikTok; 4:5 or 3:2 for Pinterest and anything print-adjacent.
Step 4: Repair instead of restarting
Most free tools now include an edit mode: inpaint, erase, extend. A stray hand or a melted mug handle is a twenty-second fix, not a reason to burn another credit on a fresh generation.
Three repairs worth learning properly:
- Inpaint: brush over the broken area, describe what belongs there (“plain wooden table surface”), and regenerate only that patch.
- Upscale: generate at the tool’s maximum size, then upscale before you crop or resize. It keeps edges clean.
- Erase: remove a distracting object and let the tool fill the gap with surrounding texture.
One hard rule: never let the AI write words inside the image. It will produce confident-looking gibberish. Generate the picture with empty space where the headline should sit, then set the type in Canva or Photopea.
Step 5: A twelve-minute run-through
Real example. Blog post titled “Why your sourdough starter smells like nail polish”. Needed a warm, appetising 16:9 header with no text.
- Prompt one: “Glass jar of bubbling sourdough starter on a floured wooden counter, morning light from a window on the right, soft shadows, shallow depth of field, 50mm editorial food photography, warm tones.”
- Problem: the bubbles were barely visible and the jar looked like plastic.
- Prompt two: added “visible bubbles, domed and active starter, glass with light condensation”. Much better.
- Finish: upscaled once, opened in Canva, dropped the headline over the left third at 30% opacity.
Twelve minutes, two generations, nothing wasted. That’s a realistic target once you’ve practised a handful of times.
What “free” usually costs you
Every free tier is free in exchange for something. Check the small print before you rely on one for paid work.
- Daily credits: typically 5 to 25 images, resetting every 24 hours. Batch your ideas in one sitting.
- Queue times: free jobs sit behind paying users. Fine for planning, painful on a deadline.
- Resolution caps: many stop at 1024px, which is fine on screen but thin in print.
- Licensing: some free plans allow personal use only. Read it before you sell anything.
- Watermarks and uploads: a few tools stamp the free plan, and some train on reference images you upload.
If you’d rather sidestep the traps entirely, this guide to getting impressive results without a subscription separates the genuinely usable free plans from credit-capped demos. Product shots are their own problem too, and the free AI photo tools built for professional-looking product images are a different shortlist from the general-purpose ones.
When a still image isn’t enough
Some projects need movement: a six-second loop for a landing page, a vertical clip for social. Free image generators stop at the freeze frame. The usual route is to generate the still here, then hand it to a video tool that turns a text prompt into a polished video in a few minutes. Two tools, one asset, still no subscription.
Start a prompt library tonight
Once you’ve generated a few good images, the temptation is to close the tab and forget how you got there. Don’t. Open a plain notes file and keep three things.
First, the full prompt text of anything that worked, including the negative prompt. Second, the seed number if the tool shows one, so you can reproduce a lucky result exactly. Third, a “house style” suffix you append to everything, something like “natural window light, muted colour palette, 50mm, subtle film grain”. That suffix is what makes a folder of images look like they came from the same brand instead of five different tools.
Fifteen minutes of note-keeping turns each new project from a guessing game into a menu. The first image you generate with a system like this will still be forgettable. The twentieth won’t be.