Here’s a collection of basic tips and tricks to get you started with generating your own AI portraits:

Start Simple. Something like “male knight with sword and shield”. You want to give the AI a solid baseline to start with, and then starting adding qualifiers to hone in on what you want.

Be prepared to iterate Don’t expect to get what you want with the first try. That almost never happens. Look carefully at what the AI did with your prompt, decide what it did wrong and what it did right, and then tweak your prompt to nudge it in the right direction. My AI Concepts: Portraits Step by Step page shows one example of what this process looks like when I go through it.

Be specific, but sparse. Describe what you want in as few words as possible, and only say the parts that are really important. The more extraneous detail you add the more it can get muddled. Every word you add should have an explicit purpose, and if the exact kind of sword she’s holding or the exact kind of belt he’s wearing don’t matter very much then let the AI make its own suggestions.

Be willing to keep clicking “Create”. If you get pics that are close, but not quite right – just run the same prompt again a few more times to see what pops out before tweaking it some more.

Be flexible. I’ve often found if I give the AI some freedom, and poke around for something that “looks cool” rather than constantly prodding for exactly what I had in mind at the beginning, then I have a better experience. I will often have one thing in mind at the beginning, but partway through am presented with something a little different that just “clicks” and I say Aah!! That’s him!!

Order Matters. The AI absolutely weighs things early in the description heavier than terms later. So I like to have the beginning of the prompt say straight up what I want. Something like “50 year old gnome couple”, followed by descriptions of the two gnomes that go “<gender> <race>, <ethnicity>, <age>, <build>,<hair>,<expression>,<pose/action>”, followed by art style tags.

Ethnicities are your friend: Using an ethnic tag early in the query – like “Italian” or “Irish” – helps solidify the look, is a great way to get a different set of faces, and also keeps your look more consistent as tweak backgrounds, poses, and action.

The AI Remembers: I can’t quite put my finger on how it works, but previous prompts absolutely influence the current prompt. If I ask for something like “orc in red studded leather with spikes”, and then ask for something like “orc in leather”, there’s still a curious preponderance of red, studs, and spikes for a few iterations. If I ask for something completely different, this tails off much faster.

Resources:

DALLE-3 Prompt Guide

SeeDream Prompt Guide:

---

AI Portrait Generation Articles

Experiments

Motion & Action

Monsters & Mayhem