The AI is not imagining your idea and then drawing it. It matches your words against millions of pictures it has already seen.
That one fact explains almost everything about what works. Words that describe what a photo looks like work well. Words that describe how it should make you feel do almost nothing.
Describe the photo, not the feeling
Words like beautiful, stunning, perfect and amazing sit next to millions of completely different pictures. So they point the AI nowhere. They feel like they are adding something. They are not.
Plain physical detail works far better. Look at the difference:
beautiful woman, amazing lighting, perfect quality
woman in her late twenties, dark hair tied back, standing by a window in afternoon light
sexy pose, gorgeous, high detail
sitting on the edge of a bed, leaning back on both hands, looking at the camera
cool background, cinematic vibe
city street at night behind her, shop signs blurry, wet road reflecting the lights
Put the important stuff first
Words near the start count for more than words near the end. Long prompts also water themselves down, because every extra thing you add competes with everything else. Twenty good words beat eighty vague ones.
Work in this order:
- 1
Who
The person, and the few details you really want kept. Rough age, hair, build.
- 2
What they are doing
Describe the body, not the mood. Leaning on a doorframe. Looking back over one shoulder. Sitting cross legged on the floor.
- 3
Where they are
The place, and what is behind them. If you leave the background out, the AI picks one for you, and it will be different every time.
- 4
How it is shot
How close the camera is, and what the light is doing. Most people skip this. It is the part that makes a picture look intentional instead of random.
Camera and light do the heavy lifting
Once you have described the person, these two things change the picture more than any extra adjectives. Pick one from each list.
How close the camera is
- Close up: head and shoulders. Use it when the face is the point.
- Mid shot: waist up. The safest choice, and good for most pictures.
- Full body: head to toe. Harder to get right, and more likely to mess up hands and feet.
- Over the shoulder: you are standing behind someone, looking past them.
- Low angle: camera down low, looking up. Makes someone look powerful.
- High angle: camera up high, looking down. Makes someone look softer.
What the light is doing
- Soft window light: gentle and flattering. Start here if you are not sure.
- Golden hour: low warm sun, long shadows. Looks great almost every time.
- Overcast: flat and even, barely any shadow. Keeps the focus on the person.
- Direct flash: harsh and bright, hard shadows. Looks like a real candid photo.
- Rim lighting: light behind them, glowing around their edge. Dramatic.
- Lamps or neon: light coming from things inside the room. Good at night.
Why you get a different face every time
Every picture starts from scratch. Nothing carries over from the last one. Run the exact same prompt twice and you get two different pictures.
So if you are trying to get the same face across a set of images, writing the description more carefully will not fix it. No amount of wording holds a face steady. That needs a saved character or a face reference, which is a different tool. If you are fighting this with words, you are fighting the wrong thing.
Things that go wrong no matter what you write
Some of this is not your fault. Knowing which is which saves you a lot of wasted tries.
- Hands. Fingers come out wrong constantly. Hands in pockets, out of frame, or holding something work far better than open hands near the face.
- Words and signs. Any text comes out as gibberish that looks almost real. Plan to add real text afterwards.
- Counting. Asking for an exact number of things rarely works.
- Left and right. These get flipped or ignored all the time. Describe the picture instead of telling it which side.
- Two people touching. Details bleed between them. One person alone is much easier.
A routine that actually gets you there
- 1
Start plain
Person, pose, place, camera, light. Nothing fancy. Get the basic shape right before you make it pretty.
- 2
Fix the big stuff first
If the pose or the framing is wrong, no amount of detail will rescue it. Sort those out first.
- 3
Then add detail
Clothes, hair, expression, small things in the background. One at a time, so you can see what each one does.
- 4
Cut what is not working
Long prompts collect dead words. If you cannot point at something in the picture, delete it. Shorter is usually stronger.
- 5
Keep what worked
Save the prompts that gave you good results and reuse the shape of them. You are building your own little library of phrases that work, and that helps more than any list of tips.
People who get good pictures consistently almost never have the longest prompts. They just worked out which few words actually matter, and stopped typing the rest.