Step 1: The Photo
Everything starts with a single 2D photo, usually taken on a smartphone chairside. The photo quality and framing matter more than most people expect — a well-lit, straight-on photo with the patient biting together and smiling naturally gives the whole pipeline a better starting point than an angled or poorly lit one.
Step 2: Detecting the Face and Teeth
Before anything gets generated, the software locates the relevant landmarks in the photo — the position of the teeth, the lip line, the midline of the face, the smile arc. This is a computer vision step, not a generative one: it's mapping what's already in the photo so the next step knows where to work and what the starting proportions actually are.
Step 3: Applying Clinical Guardrails
This is the step most generic AI photo editors skip entirely, and it's the difference between a believable simulation and an obviously fake one. The detected landmarks get checked against the same things a trained eye would check — smile arc, occlusal cant, gingival display, tooth proportion, arch form. (We cover these in more depth in our realism guide.) These aren't abstract rules; they're the specific measurements that separate a natural-looking result from a "denture ad" one.
Step 4: Generating the Image
With the landmarks mapped and the clinical constraints set, the generative step produces the actual simulated image — the teeth as they'd realistically appear with the treatment applied, respecting the face's proportions rather than pasting on a generic "perfect smile." For video simulations, this extends to generating the patient smiling and talking with the new result, not just a static frame.
Step 5: Dentist Review
The generated result isn't shown to the patient automatically. The dentist reviews it first — checking that it's realistic for the case and appropriate to show. This step matters clinically (catching anything that looks off before the patient sees it) and for trust (a simulation the dentist has personally reviewed carries more credibility than one generated and shown unsupervised — we cover why in our patient trust guide).
Why "Just Upload a Photo" Undersells What's Happening
From the user's side, the process looks like: take a photo, wait 30 seconds, get a result. That simplicity is the point — nobody wants a 10-step manual workflow chairside. But "simple to use" and "unsophisticated" aren't the same thing. The landmark detection and clinical guardrail steps are what make a 30-second simulation clinically credible instead of an AI photo filter with teeth whitened.
What This Process Doesn't Do
It doesn't diagnose anything, doesn't replace a clinical exam, and doesn't produce a manufacturing file or treatment plan — that's a different category of software entirely (see our comparison of smile simulation vs digital smile design). It generates a visual preview from a photo. That's the whole job, and it's a narrower job than the phrase "AI-powered" sometimes implies.