Your first 10 AI images are going to be bad and you're going to assume the tool is bad.
The tool is fine. You typed 6 words and expected it to read your mind.
Here's the fix, and it really is this simple: an image prompt needs 5 things. Leave one out and you get that plasticky, over-lit, obviously-AI look everybody recognizes now.
The 5 things
1. What it is. The subject, specifically. Not "a man." A 40 year old man with a short beard in a navy work shirt.
2. What it's doing / where it is. Standing where? Holding what? Sitting at what?
3. The light. This is the one everybody skips and it's the one that makes an image look real. Soft window light from the left. Overcast afternoon. Harsh midday sun. Warm lamp light at night.
4. The shot. How close, from where. Close-up, waist-up, wide. Eye level, from slightly above, from the floor.
5. The look. Photo or illustration? Grainy or clean? What kind of camera feel?
Miss the light and it looks fake. Miss the shot and you get a random crop. Miss the look and you get the default AI shine.
Watch the difference
6 words:
a coffee cup on a tableYou get a stock-photo-looking mug, lit from nowhere, floating on a beige surface.
The 5 things:
A white ceramic coffee cup on a worn wooden table, steam
rising. Soft morning light from a window on the left, long
shadows. Shot from slightly above at a 45 degree angle,
close-up. Natural photo, shallow depth of field, slight grain,
not glossy.Same subject. One of them looks like a photograph somebody took.
The 3 you'll actually use
A headshot that doesn't look AI
Upload a real photo of yourself first, then:
Using this photo as the person, create a professional headshot.
Same face, same hair, keep it recognizably me.
Wearing [what you'd actually wear to meet a client].
Plain background, soft grey.
Lit from the front left with a large soft light, gentle shadow
on the right side of the face.
Waist up, shot at eye level, 85mm lens look.
Natural skin texture. Do not smooth or retouch the skin.That last line is the whole trick. Left out, every AI headshot comes back looking like a wax figure. "Do not smooth the skin" is the difference between usable and uncanny.
A product shot
A [your product] on a [surface], photographed for a website.
Lit with one large soft light from the upper left, and a subtle
fill on the right so the shadows aren't black.
Shot straight on at product height, centered, plenty of empty
space around it.
Clean commercial photo. Sharp on the product, background
slightly out of focus. No text, no logos, no props competing
with it.A thumbnail or ad background
[Describe the scene], leaving the entire right third of the
frame empty and simple so text can go there.
High contrast, bright, one dominant color.
Shot wide.
Bold and graphic, reads clearly at small size.That empty-third instruction is why some thumbnails look designed and others look like a photo with words dumped on top.
Rest of it's yours
Drop your email and the whole thing opens up right here. No waiting on a confirmation, no download, nothing to install.
You'll also get the new ones as I make them. Leave whenever, no hard feelings.
Fixing an image instead of rerolling it
The mistake everybody makes: image is 80% right, so they generate again from scratch and get something entirely different.
Don't reroll. Adjust:
Keep everything the same. Change only the light: make it
warmer and lower, like late afternoon instead of midday.Same image, but pull the camera back so I can see the whole
room.That's right but too perfect. Add some wear. Scuffs on the
table, slightly messy background, like a real place people use."Too perfect, make it real" is the single most useful correction in AI imagery. Almost every bad AI image is bad because it's too clean.
The tells that give it away
If somebody looks at your image and instantly knows, it's usually one of these:
- Everything is lit from everywhere. Real light comes from somewhere. Say where.
- Skin like plastic. Always ask for natural texture, always say don't retouch.
- Impossible tidiness. Real desks have a cable. Real tables have a mark.
- Dead center, perfectly symmetrical. Ask for it slightly off-center.
- Text. AI still mangles text inside images. Don't ask it to write words in the picture. Generate the image clean, put the text on afterwards.
That last one matters if you're making thumbnails or ads. Leave the space, add the words yourself in any editor. Every time.
When you need the same look twice
For a brand you need consistency, and rerolling won't give it to you. Write your look down once and reuse it:
My image style, use this every time:
Natural photography, never glossy or 3D-rendered.
Soft directional light, one main source.
Muted colors, slightly desaturated.
Real texture, a little grain, nothing perfect.
Plenty of empty space in the frame.
Shot at eye level.Paste that at the top, then describe the specific image underneath. Now 40 images look like they came from one place.
If you're doing this constantly, put it in a skill so you never paste it again. That's here.
The honest limits
It's genuinely good at: backgrounds, product shots, moods, textures, illustration, anything where "close enough" is fine.
It's still bad at: text in the image, hands doing something complicated, keeping a specific real person perfectly consistent across many images, and anything that has to be factually exact.
And the part worth saying out loud: if you need a photo of your actual product to sell your actual product, take the photo. A generated stand-in of a thing you really sell is how you end up with a customer holding something that doesn't match the picture.
Everything around it though, backgrounds, moods, ads, thumbnails, the stuff you were paying for or skipping entirely, that's yours now for the price of describing it properly.
Go describe one thing with all 5 parts. You'll see it immediately.