How AI headshots actually work.

Under the hood of every AI headshot tool is the same exchange: the model learns your face from example photos, then places that face into a newly generated image. The quality gap between tools — and between good and bad results on the same tool — comes down to how carefully those two steps are handled. Here's the whole pipeline, demystified.

01

Step one: the reference set

Everything starts with 2–6 photos of your face. The model reads them to build an internal understanding of your features — bone structure, skin, hair, the way your face changes across angles. This is why set quality dominates results: six identical selfies teach one angle; varied, sharp, well-lit frames teach a face.

Darkroom grades your set before you spend anything. The set check scores every frame for light, angle, sharpness and cross-photo consistency, and names the weakest frame so you can replace it. It's unlimited on every plan because a good set is the cheapest insurance there is.

02

Step two: the brief

The second input is the shot itself. In Darkroom, that's either a pre-lit pack — a complete setup of key light, backdrop, lens and styling — or your own description, which the AI expands into a photographer's brief: lighting ratios, camera body and lens, pose, wardrobe, colour story. The ✨ AI Prompt Writer can draft that brief from a single plain-English sentence.

03

Step three: the render

The model then generates a brand-new image: your face, reconstructed from the references, placed inside the brief's lighting, lens and setting. Nothing is pasted or retouched — it's a fresh render, which is why the light wraps naturally and why it takes about 15 seconds rather than 15 milliseconds. Identity preservation is the hard part of this step, and it's why we built identity-first prompting: the brief tells the model to protect your features before it chases the style.

  • Your likeness comes only from your reference photos
  • Light, lens, pose and wardrobe come from the brief
  • The image is generated fresh — no compositing, no filters
04

Why some AI headshots look fake

Almost every obvious AI headshot fails in one of three places: a weak reference set (so the face drifts), an overreaching brief (so the face is fighting the style), or plastic over-smoothing in the model. The fixes are the set check, honest briefs, and treating renders as look tests — which is exactly how the studio is built.

Questions

Asked and answered.

Does the AI train a permanent model of my face?

No. Your reference photos stay in your browser, are sent to the model only for the check or render you trigger, and are never stored on our servers. There's no standing model of you to delete later.

Why do I need several photos instead of one?

One photo gives the model a single angle and expression — renders then drift as soon as the brief asks for a different pose. 2–6 varied photos let it reconstruct your face in three dimensions, which is what keeps the likeness stable across styles.

How long does a render take?

About 15 seconds. The set check and shot briefs are near-instant, unlimited, and cost nothing on any plan.

What does a generation cost?

A standard look test costs 1 credit; a ✨ Darkroom Finish premium pass costs 2. Free accounts start with 5 credits, no card. Pro is A$20/month for 100 credits, and top-ups add 25 for A$4, 50 for A$7 or 100 for A$12.

Keep reading

See it on your own face.

Five free credits, no card. Run the set check, load a pack, and render your first look test in about 15 seconds.