Face expressions you actually want
Smile the way you smile in real life. Mix in two or three serious frames so the model has range. The expressions you upload are the expressions you receive.
Selfie guide
10 selfies, 3 minutes, the cleanest possible AI headshots
Aurawave's Intelligence Engine grades every output. The grading is only as good as the selfies it has to work with. Follow this guide and the grade gets cleaner, the regenerations get rarer, and the delivered set looks more like you.
What works
Examples of good photos
Six rules. The model trains on these directly. Get them right and the grading checklist has good material to work with from the first generation.
Large, high-resolution face
Aim for a face larger than 1024px on the long edge. Phone-camera selfies at arm length easily clear that bar. Cropping to a tiny face leaves the model nothing to learn from.
Face is the focus
Centered, sharp, well-exposed face. The background can vary, but the face cannot be the secondary subject. If you have to squint to find your face, the model will too.
Recent photos taken with a real camera
Phone selfies from the last 6 months work. Old photos where you looked different lower the identity match score on every output the engine grades.
Range of outfits and backgrounds
Two or three different shirts, two or three different rooms or outdoor settings. Variety teaches the model what is you and what is just a coincidence of one shirt.
Clear face in good lighting
Window light is best. Overhead office light works. Avoid harsh shadows across the face and avoid backlight that turns your face into a silhouette.
What doesn't work
Photos that won't work
Avoid these. The Intelligence Engine catches a lot of bad output, but it can't fix bad input. These selfies leave the model nothing usable to train on.
Heavy face filters
Filtered selfies make the AI output look extra-plastic. The filter trains the model on the filter, not your face.
Hats, sunglasses, AirPods, headphones
Anything covering or distorting the head shape interferes with the identity grade.
Black-and-white or sepia
Color information matters. The grading layer evaluates skin tone match, and it cannot evaluate what was stripped out.
AI-generated photos as input
Do not feed the model its own outputs. The drift compounds and the identity grade collapses by generation 2.
Face cut off or partially hidden
The model needs the full face from forehead to chin. Cropped, hands over face, behind objects, all out.
Wedding or formal-event photos
Heavy makeup, formal hair, and unusual lighting from one event do not generalize. Save the wedding album for the wedding album.
Multiple people in the frame
The model gets confused about which face to learn. Solo selfies only.
Blurry, grainy, or tiny face
Low resolution or out-of-focus faces are unrecoverable training data. The grade caps at the input quality.
Goofy expressions you do not want
Tongue out, eyes crossed, mid-laugh frames. Whatever you upload, you will receive variants of.
Distorted angle from extreme selfie
Phone too close, looking up, looking down. The face geometry warps and trains the model on the warp.
3-minute at-home photoshoot
Side by side, what to do and what to avoid
Two columns. The left is what the Intelligence Engine wants to see. The right is what trips it up.
Recommended
-
Variety of expressions
-
Half-body and full-body mix
-
High-quality images
-
Face in focus
-
Hair styled the way you want
-
Good lighting
Not allowed
-
Multiple people in photo
-
All full-body, no face detail
-
Heavy filters or face-tune
-
Face cropped off
-
Goofy or extreme expressions
-
Bad lighting, blurry, or grainy
The grade
What happens after you upload
-
01
Generation
The model trains on your 10 selfies and produces a candidate set of headshots across the attire and background combinations you selected.
-
02
Grading
The Intelligence Engine scores every output against a working photographer's checklist: catchlight, fill ratio, sharpness, jaw separation, identity match.
-
03
Delivery
Failures are killed and regenerated until the set is clean. You receive about 25 graded photos in under 90 minutes.
A note on refunds. Uploaded selfies that ignore this guide are not eligible for a redo. Read the refund policy before uploading if you want the full picture.
Ready? 10 selfies, 3 minutes.
Upload your selfies and the Intelligence Engine takes it from there. You'll have your graded set in under 90 minutes.