The short version
- You can make a good AI image for free, today, without installing anything.
The free tools are genuinely usable — they limit how many you make, not how good they look. - Describe the picture, not the idea. “A tabby cat asleep on a windowsill,
afternoon light, shot on film” beats “a nice cat photo” every time. - Check the licence before you use one for business. Free tiers frequently
allow personal use only, and that restriction is easy to miss. - Paid only starts to matter when you need volume, higher resolution, or the
commercial rights.
Making an image with AI has stopped being a technical skill. You type a description, wait
fifteen seconds, and get a picture. This guide is for someone doing that for the first time —
what to use, what to type, and the two or three things that trip people up.
How do you make an AI image?
You open a tool, type a description of the picture you want, and it generates one. That is the
whole process. The difference between a disappointing result and a good one is almost entirely in
how you describe it — not in which tool you picked, and not in any setting you need to
understand.
Every mainstream tool now works in a browser with a free account. You do not need a powerful
computer, because the work happens on the company’s servers rather than yours.
Write the description like a photographer, not like a wish
This is the one skill worth learning, and it takes about five minutes.
A weak prompt names a subject: “a coffee shop”. A strong prompt names the subject and
then the things a photographer would decide — the light, the angle, the mood, the medium:
“A small coffee shop interior at 7am, morning light through a large window,
steam rising from a cup on a wooden counter, warm tones, shallow depth of field”
Four ingredients cover most of it:
- Subject — what is in the picture, specifically. “Golden retriever puppy”, not
“a dog”. - Setting — where it is and what time of day.
- Light — soft, harsh, golden hour, neon, overcast. This changes the result
more than anything else on the list. - Style — photograph, watercolour, 3D render, pencil sketch, oil painting.
Say what you do want rather than what you don’t. Asking for “no people” often produces
people, because the model reads the word “people”. Describe the empty street instead.
What free actually means
Free tiers in this category are real, but they are promotional — they exist to move you onto a
paid plan, and their terms shift. The pattern is consistent even when the numbers are not:
- A capped allowance, either a daily quota or a one-off deposit of credits.
Runway, for example, gives 125
one-time credits that do not renew, with paid plans starting at $12 a month. - A watermark on the output, or a lower resolution than paying users get.
- A slower queue when the service is busy.
- Personal use only. This is the one that catches people out. A free plan often
grants no commercial licence at all, so an image you generated for a client or a product listing
may not be yours to use that way.
Read the licence line on the pricing page before you use anything commercially. It is usually
one sentence, and it is usually the difference between free and not.
When is it worth paying?
Three situations, and only three:
- You need the commercial rights. Any business use — a website, an ad, a
product — makes this the whole question. - You are making a lot. Free quotas are sized for trying things, not for
producing forty images a week. - You need the resolution. Free output is often fine on a screen and too small
for print.
If none of those apply, free is genuinely enough. Most people never need more.
The mistakes that waste the most time
- Giving up after one attempt. Generation is partly luck. The same prompt run
four times gives four different pictures, and the fourth is often the one. - Writing a paragraph. Long prompts dilute. Twenty well-chosen words beat
eighty. - Expecting text to work. Words inside an image — a shop sign, a book cover —
are still where these models fail most visibly. Add text afterwards in any editor. - Expecting hands to work. Better than they were, still the first place a
picture goes wrong. Ask for a pose where hands are not the focus. - Fighting the tool instead of restating. If three attempts miss, the
description is the problem, not the model.
One prompt, rewritten
The gap between a weak prompt and a strong one is easier to see than to explain. Same picture,
three drafts:
| Draft | Prompt | What you get |
|---|---|---|
| First try | a woman working on a laptop | A generic stock image. Nothing decided, so the model decides everything. |
| Better | a woman working on a laptop in a cafe, morning | Recognisable scene, still flat — no light, no distance, no style. |
| Working | a woman in her thirties working on a laptop at a window seat in a cafe, early morning light from the left, shot from across the table, shallow depth of field, documentary photography |
A specific photograph. Every choice a photographer makes is now yours. |
The third is not longer for its own sake. Each clause removes one decision from the model and
gives it to you. That is the entire skill.
Fixing a result instead of starting over
When an image is close but wrong, change one thing rather than rewriting from scratch.
- Wrong mood? Change the light. “Overcast” and “golden hour” produce two
different emotions from an identical scene. - Too cluttered? Describe the background explicitly instead of leaving it
open — “plain grey wall behind” rather than hoping for one. - Wrong framing? Say the distance: close-up, waist-up, wide shot.
- Looks like a stock photo? Name a medium. “35mm film”, “watercolour”,
“pencil sketch” pull it away from the polished default.
Keep the version that nearly worked. Regenerating a prompt you have already improved is how
people spend an hour and end up back where they started.
Frequently asked questions
Do I need to pay to make AI images?
No. Every major tool has a free tier that produces the same quality as the paid one — what you
get less of is quantity, resolution and, importantly, commercial rights.
Can I sell something made with an AI image?
Only if the plan you used grants a commercial licence, and free plans frequently do not. Check
the pricing page for that specific line before using an image for business.
Why do the faces and hands look wrong?
Hands and text are the two hardest things for image models, because both need exact structure
rather than a plausible texture. Generate several attempts, or compose the shot so neither is the
focus.
Do I need a good computer?
No. Browser tools run on the provider’s hardware. A phone is enough. Only running a model
locally on your own machine needs a capable GPU.
How long does one image take?
Usually seconds. Free tiers put you in a slower queue when the service is busy, which is often
the most noticeable difference between free and paid.
Next: animate a still into video · build a faceless reel around it
