What to wear for your AI headshot training photos
Most people think the AI generator only learns their face. It also learns their clothes. That is fine when you are generating headshots in the same outfit you uploaded. It is a problem when you ask for a tailored blazer and the model keeps generating you in the t-shirt you wore in eight of your twelve training photos.
The rules below give the model enough visual room to separate "you" from "your clothes," and let you generate outfits you did not actually upload.
Solid colors over patterns
Solid colors are easier for the model to ignore than complex patterns. A plain gray t-shirt, a solid navy sweater, a cream blouse. These become visual background. A shirt with a big logo, a loud plaid, or a graphic print becomes something the model tries to learn as part of you.
Bring variety
Do not upload twelve selfies in the same shirt. Even if it is your favorite shirt. Vary the clothing across photos so the model understands that clothing changes while you stay the same. A practical mix looks like this.
- Two or three in something formal, like a blazer or a collared shirt.
- Two or three in something casual, like a plain t-shirt or sweater.
- A couple in something in between, like a knit polo or a simple dress.
- If you wear glasses daily, include a couple with them on and a couple without.
What to avoid
- Hats. The model will either bake a hat into every output, or struggle with your hairline.
- Sunglasses. Eyes are the most important signal for training. Cover them and the model underperforms.
- Hoodies with the hood up. Same problem as hats.
- Shirts with large text or logos. Those end up rendered as gibberish in outputs.
- Heavy makeup or heavy filters. The model will learn those as part of your baseline appearance.
Do I need to dress up
No. The output style is controlled by the prompt, not by what you wore in training. You can upload twelve photos in plain t-shirts and generate a set of executive portraits in tailored suits. The model learns what your face, hair, and build look like. The rest is rendered fresh at generation time.
The ten-minute outfit check
Before you upload, glance through the selection once and ask these questions.
- Is the same shirt in more than half the photos?
- Is anything covering my hairline, eyes, or lower face in more than two photos?
- Are there bold patterns or logos in more than two photos?
- Is the overall palette extremely similar across all photos?
If you answered yes to any of these, swap a few photos before you run training. Five minutes of curation at this step saves an hour of frustration later.
Frequently asked.
- Can I wear glasses in my training photos?
- Yes, if you wear them every day. Mix some with and some without if you use them only sometimes. This lets the model render both versions depending on the prompt.
- Does my hair need to be styled the same way in every photo?
- No, and variety actually helps. Different hair in different photos teaches the model what stays the same about your face across hairstyles.
- What if I want headshots in a suit but I never wear one?
- That is fine. The model learns your face and build from the training photos, then generates new wardrobe at prompt time. You can train in hoodies and generate tuxedos.