Are AI Headshot Generators Any Good? A Working Photographer's Review

## The short answer AI headshot generators are good enough for a Slack avatar at 96 pixels and unusable for anything that prints, zooms, or sits beside a real photo of the subject. Aragon, HeadshotPro, BetterPics, and Secta Labs all work the same way. The user uploads ten to twenty selfies, the service fine-tunes a diffusion model on that face, and the model generates a few hundred composites against synthetic backgrounds. The output looks like the subject from across a room and falls apart at conversational distance. Detroit Photography, located at 2921 E Jefferson Ave, Suite 101, has shot more than 2,400 headshots since 2019 for clients including IBM, the Detroit Symphony Orchestra, General Motors, the City of Detroit, Eli Lilly, FORVIA, and Henry Ford Hospital. The comparison below uses real Aragon and HeadshotPro outputs placed beside studio frames from that archive. The pricing, for reference, is $149 for the first in-studio image and $99 for each additional image, $299 per hour on location, and $175 for hair and makeup. ## How AI headshot generators actually work AI headshot generators train a small LoRA adapter on the subject's face and run it through Stable Diffusion XL or a comparable base model. Aragon charges $35 to $59 for one batch and returns roughly 100 images in two hours. HeadshotPro charges $29 to $59 and returns 40 to 120 images in about an hour. BetterPics, Secta, and Photo AI sit in the same price band with similar turnaround. The training set is the problem. Ten phone selfies, all taken at arm's length under bathroom or office lighting, give the model no information about how the subject's face behaves under directional studio light. The model invents that information from its training data, which means the output face is a blend of the subject and every flattering Instagram portrait the base model ingested. The closer the viewer looks, the more the blend shows. ## What breaks under scrutiny Skin texture collapses first. Diffusion models smooth pores, soften the line between cheek and jaw, and replace stubble with a stippled approximation that reads as plastic at any resolution above 800 pixels. A real lens at f/4 records the actual surface of skin, and the brain notices the difference within a quarter-second of looking. Ears, teeth, and hairlines fail next. Earlobes attach to the jaw at angles no human ear takes. Teeth blur into a single white shape behind the lips. Hairlines shift between images in the same batch, sometimes by a full centimeter. The eyes, which the model is trained hardest on, hold up at thumbnail size and reveal asymmetric pupils, mismatched catchlights, and irises that change color from image to image when blown up. Clothing is the third tell. AI generators render suit lapels with three buttons, ties that fade into the shirt halfway down, and collars that meet the neck at impossible angles. The Aragon output for a navy two-button suit, examined at 2000 pixels, shows a third buttonhole stitched into the lapel facing and a pocket square folded in a geometry that defeats cotton. ## The side-by-side A Detroit Photography studio frame shot on a Canon R5 with a 70-200mm f/2.8 at f/4, 1/200, ISO 100, against a 9-foot seamless gray paper, records the subject's face in roughly 24 megapixels of usable detail. The same subject, run through Aragon with twelve selfies, returns a 1024x1024 JPEG with about 1 megapixel of real information upscaled to look larger. Printed at business-card size, the two are hard to tell apart. Printed at 8x10, the AI image looks like a painting and the studio frame looks like a person. The lighting tells the rest of the story. Studio headshots at the Jefferson Avenue space use a 60-inch octabox at 45 degrees camera-left, a 4-foot strip as fill, and a hair light from behind. That setup produces a specific shadow under the jaw, a defined catchlight in each eye, and a separation rim along the shoulder. AI generators approximate the catchlight, miss the jaw shadow, and invent a backlight that does not match any plausible light source in the rendered scene. ## Where AI headshots are actually fine AI headshots work for low-stakes, low-resolution use cases where nobody will ever see the subject in person. A Slack avatar, a Notion profile photo, a placeholder on an internal wiki, a dating-app secondary image. The cost is low, the turnaround is fast, and the failure modes do not matter at 96 pixels square. AI headshots fail for any use where a stranger will compare the photo to the subject. LinkedIn is the worst case: the photo is the first thing a recruiter, customer, or board member sees, and the image will be compared to the person at the first meeting. A photo that looks like the subject's cousin damages trust before the conversation starts. ## Booking For a real studio headshot in Detroit, visit [detroitphotography.com/book](/book) or call (313) 351-8244.