Every preview is checked before you show it
A second AI model grades each smile preview against the patient's own photo: did it keep their face, their framing, and the treatment you asked for? You see the answer first, and a weak preview can be regenerated with the fix already applied.
The short version
The whole reason a smile preview closes a case is that the patient recognizes themselves in it. The moment an image quietly widens the crop, softens the lips, or comes back a shade brighter than anyone asked for, it stops being a preview of their smile and starts being a stock photo with their face on it. Patients feel that even when they cannot name it.
So Smile PreVue checks its own work. Every finished preview is handed to a separate AI model along with the original photograph and the exact instructions used to generate it. That model grades seven specific things and returns a Fidelity Score from 0 to 100, shown on the patient record as Excellent, Good, or Review.
It is a check for the clinician, not a badge for the patient. The score never appears on a share link, a PDF, or the payment page. It exists so that the one image you put in front of someone deciding on thousands of dollars of dentistry is one you have already vetted.
How the check works
The preview is generated
You photograph the patient, pick the treatment and the VITA shade, and the preview comes back in about 30 seconds. Nothing about the check slows that down.
A second model grades it
In the background, a separate AI model compares the preview against the original photo and the exact instructions used to make it. It looks only at the teeth, the framing, and whether anything else was left alone.
You see the result first
The Fidelity Score lands on the patient record as Excellent, Good, or Review, with a plain-language caption. It is a chairside check for the clinician, not something the patient ever sees.
Regenerate with a targeted fix
If a preview lands in Review, Smile PreVue names what went wrong and offers to regenerate with that specific correction applied. The earlier attempt is archived, never deleted.
The seven things it grades
Each one is scored individually and weighted into the final number, so a preview that nails the shade but reframes the shot does not get to hide behind an average.
Identity preserved
The face, skin, eyes, hair, expression, pose, and background must match the original photo. The only thing allowed to change is the teeth.
Framing preserved
Same crop, same zoom, same camera position. A preview that quietly reframes the shot is not a fair before-and-after.
Lips untouched
Lip shape, position, color, and texture stay as they were. The smile line belongs to the patient, not to the model.
Teeth-only edit
Every change is confined to tooth surfaces that were already visible. No teeth invented, revealed, or removed.
Treatment fidelity
The change matches the treatment you actually requested, and nothing the instructions ruled out was done anyway.
Shade match
Tooth color matches the VITA shade you selected, with a natural gradient from the gumline to the edge. Not graded on whitening or alignment cases, where no shade is specified.
Realism
Photorealistic enamel with natural translucency, texture, and lighting. No artifacts, no blur, no glow.
A bad preview should cost you one tap, not the case
When a preview lands in Review, guessing at what to change is the slowest possible fix. Smile PreVue instead tells you what the grader caught, in a sentence you can act on: the crop moved, the shade overshot into an opaque white, the shape changed on a whitening case.
From there you have two buttons. Regenerate with fix runs the preview again with that specific correction written into the instructions. Regenerate as is simply tries again. Either way the previous attempt is archived rather than discarded, so the patient record keeps the full history and you can compare attempts side by side and return to an earlier one.
In practice that turns the awkward chairside moment, where you are quietly deciding whether an image is good enough to show, into a decision the software has already made visible.
A quality check, not a clinical claim
The Fidelity Score measures one thing: how faithfully the generated image kept the patient's face, the framing, and the treatment rules you specified. It does not diagnose, it does not plan treatment, and it does not predict how the finished restoration will look. A simulation is a communication tool, and the clinician decides what is clinically achievable.
Grading runs on the same BAA-covered Google Cloud infrastructure as the preview itself. More detail on the security and compliance page, and everything that is included on the pricing page.
Fidelity Score FAQ
What is the Fidelity Score?
It is an automatic quality check on every AI smile preview. A second AI model compares the finished preview against the patient's original photo and the instructions used to generate it, grades seven specific things, and returns a single score from 0 to 100 shown as Excellent, Good, or Review. It is included in your subscription at no extra cost.
Does the patient see the score?
No. The score appears only inside your dashboard, on the patient record, for your signed-in team. It is not on the share link you text or email, not on the PDF, and not on the payment page. The point is that you see a weak preview before the patient does.
Does it predict how the finished case will look?
No, and this is an important distinction. The Fidelity Score measures how faithfully the preview kept the patient's own face, framing, and the treatment rules you specified. It is a check on the image, not a clinical prediction, not a diagnosis, and not a promise about the final restoration. Your clinical judgment decides what is achievable.
What happens when a preview scores low?
It is flagged as Review and Smile PreVue names the specific problem, for example that the crop changed or the shade came back more opaque than you asked for. You can regenerate with that correction applied automatically, regenerate as is, or keep what you have. Nothing is thrown away without you.
What happens to the previous attempt?
It is archived, not deleted. Every attempt stays on the patient record under Previous attempts, with its own score, so you can compare them and go back. Share links that were already sent point at the attempt they were created from.
Does the check slow down the preview?
No. Grading runs in the background after the preview is already on screen. You can present the case immediately; the score catches up on the patient record.
Does the grader see or describe the patient?
It is explicitly instructed to grade only the teeth, the framing, and whether non-dental pixels were left unchanged, and never to describe the person's identity or appearance. Grading runs on the same BAA-covered Google Cloud infrastructure as the preview itself, and the patient photo never leaves it.
Can I turn it off?
Yes. The score card, the attempt history, and the regenerate buttons can be switched off for your practice if you would rather not see them. Contact support and we will turn the display off for your clinic.
Keep reading
Where AI belongs in a dental practice, and where the dentist stays in charge.
Show the preview you already checked.
Try it on your next consult. 3-day free trial, no card, no contract.