Most people type something like "a portrait of a woman" and then wonder why the result looks flat. The tool isn't broken. The prompt is just too thin for it to work with. Getting a good portrait out of any AI image tool comes down to four or five details, and once you know them you can reuse the same structure over and over. If you want to try things as you read, the AI image generation tool on the homepage is a good place to experiment.
Start with the subject and the framing
Be specific about who is in the shot and how close the camera is. "A woman" is vague. "A woman in her late thirties, close-up head and shoulders" gives the model a clear job. Framing words like close-up, medium shot, full body, and wide shot change the composition far more than most people expect. Pick one and stick with it instead of stacking three different framings into a single prompt.
Name the light
Lighting is the single biggest lever for realism. Vague prompts produce the flat, evenly lit look that screams artificial. Try terms like golden hour, soft window light, rim light, overcast daylight, or hard noon sun. If you want drama, add a direction: side lit, backlit, light from the left. Descriptive lighting almost always beats adjectives like "beautiful" or "stunning," which tell the model nothing it can render.
Add camera language
Portrait photographers talk in focal lengths for a reason. Mentioning an 85mm lens with a shallow depth of field gives you the blurred background and gentle compression that reads as a real photograph. Wider angles like 24mm push the face and distort features, which is fine if you want energy, and wrong if you want a flattering headshot. Lens and aperture words are short and they do a lot of heavy lifting.
Say what you don't want
It helps to state the things that ruin portraits: extra fingers, warped eyes, plastic skin, text overlays, watermarks. Many tools now accept a negative prompt field. Even without one, you can often fold a short phrase like "natural skin texture" into the main prompt and get a cleaner result.
Three mistakes that keep showing up
The first is piling on conflicting styles. "Photorealistic anime oil painting" gives you mush. Pick one direction. The second is writing a paragraph of backstory. The model doesn't need her childhood or her job, it needs visual information. The third is accepting the first output. Run the same prompt three or four times. The wording didn't change, but the seed did, and the difference is often night and day.
A prompt template you can steal
Subject and age, expression, framing, location, lighting, lens, mood. Something like: "A man in his fifties, gentle smile, head and shoulders, standing in a kitchen, warm window light from the left, 85mm lens, shallow depth of field, calm mood." That's it. Fill in the slots and you have a prompt that will beat ninety percent of what people type.
Practice beats theory
Prompt writing is a hands-on skill. Reading about lighting won't help until you've seen how "backlit" versus "side lit" changes a face. Spend twenty minutes testing one variable at a time and the vocabulary starts to stick. Head back to the text to image generator and try the template above with your own subject. Change one word per run, and you'll learn faster than any tutorial can teach you.