5 Creative Ways I Use the Latest ChatGPT Image Model for Personal Portraits

· 2 min read
Two women pose confidently in professional attire in a studio setting.

Photo by Airam Dato-on on Pexels

The default thing people ask AI image models for is some version of “portrait of me as a [cyberpunk hacker / Renaissance noble / Studio Ghibli character].” The output looks fine the first three times and forgettable forever after.

Here are five patterns I’ve been using that produce portraits I actually want to keep.

1. Style cross-pollination instead of style copying

Instead of “in the style of Wes Anderson,” try:

Portrait, color palette of Wes Anderson’s Grand Budapest, lighting of a 1970s Polaroid, composition of a Vogue cover from the 90s.

Three references, none of them dominant. The result feels more like a new thing than a copy.

2. Mood over costume

Specifying the emotional state gets you a better picture than specifying the outfit. Compare:

  • Bad: “portrait of me in a leather jacket”
  • Better: “portrait of me, exhausted but quietly proud after finishing something hard”

Costume direction is shallow. Mood direction makes the model commit to something real.

3. Environment storytelling

The space around the subject does more work than the subject’s pose. “Portrait of me in a workshop where I clearly spend a lot of time, lit only by a desk lamp at 11pm” produces a vastly better image than “portrait of me looking serious.”

Tell the model what was happening just before the photo. It’ll fill the frame accordingly.

4. Consistent self as a character

Once you find a generation you love, treat the description of that face as your reference. Re-use it across new scenes, lighting setups, and outfits. You’ll build a small library of portraits that look like you — recognisably the same character — in many contexts. It’s the closest thing to having your own actor.

5. Reference photo + redirection

Upload an actual photo of yourself and say:

Keep the face and proportions exactly as in the reference. Change the lighting to [X]. Change the environment to [Y]. Style the post-processing like [Z].

This produces the most usable results by a long way. Real reference, controlled edits. It also stops the model from “improving” your face into someone else’s.

The pattern that almost never works

“Make it look professional.”

The model has no idea what “professional” means in your context. It’ll default to a stock-photo headshot every time. If you want professional, give it specifics: lighting, background, composition, camera angle. Otherwise it’ll guess — and guess generically.

The model is a creative tool, not a vending machine. Treat it like a collaborator with strong instincts and short attention.