AI Photo Analysis Is Moving From Skin Care to Cosmetic Advice

George Mack described a two-year skin condition that resisted both medical advice and lifestyle experimentation before Gemini produced the practical suggestion that worked for him.

Mack said his face would break out red enough that he did not want to leave the house. He had spoken to “a few different doctors,” many of whom recommended topical steroid creams or attributed the problem to stress. That sent him deep into meditation and into dietary restriction; he said he cut sugar so aggressively that he once received a false diagnosis of type 1 diabetes.

The eventual fix, he said, came when he uploaded years of photos to Gemini. Chris Williamson clarified whether Mack meant a single photo; Mack said no, he had uploaded “all the photos” he had tracked over time. Gemini’s recommendation was simple: put Nizoral shampoo on his face. Mack said the issue has not returned since, and that he now uses it once every two weeks.

Tim Ferriss immediately inferred the likely mechanism: “So it was fungal?” He added that Nizoral is ketoconazole and, for listeners unfamiliar with it, can be effective for topical fungal infections. Nirav Savjani noted that it is commonly used for dandruff.

Mack’s practical recommendation was broad, built from his own case: upload photos of your skin to Gemini or ChatGPT and ask for moisturizer and care recommendations based on what the model can see. In his account, the model produced a useful suggestion after years of tracking, doctor visits, stress explanations, and topical-steroid recommendations had not resolved the issue.

Nirav Savjani brought the photo-upload use case back from skin care to appearance. He said people are using ChatGPT, Gemini, and app-store wrapper products for “looksmaxing”: uploading sets of photos and receiving suggestions about how to improve their appearance.

The suggestions, as Savjani described them, can range from medical procedures and jaw surgery to symmetry, hairstyle, or beard choices. The underlying use case is simple: a model looks at a person’s face and recommends changes. Savjani called it a “huge use case for AI” because it meets an existing demand around appearance optimization.

Chris Williamson cited Qoves as an example. He described the site as offering a “glow up without surgery” through personalized facial analysis and transformation plans based on 2,000 academic studies. He said he knew the science team behind it and described them as “absolutely sick.”

The qoves.com page shown on screen displayed a before-and-after interactive face slider and a list of promised benefits: “Get more career opportunities,” “Boost your self-confidence,” “Make a stronger first impression,” “Improve your dating life,” and “Enhance your quality of life.” The same screen included publication names such as Daily Mail, Business Insider, Cosmopolitan, and GQ.

The on-screen framing connected facial analysis to career opportunities, confidence, first impressions, dating, and quality of life. It was not only a claim about looking better in the abstract; the page itself presented appearance optimization as tied to social and practical outcomes.

The slider drew mixed reactions. Ferriss characterized one transformation as “drive-by shooting suspect to soap opera star.” When the slider was moved across a woman’s face, he compared it to the familiar movie trope where a supposedly unattractive character becomes attractive by taking off glasses and changing her hair. Williamson pushed back slightly, noting that the face had also been shaped.

Savjani framed the broader category as “IRL Facetune,” while clarifying that Facetune itself manipulates photos rather than the person. He described it as a long-standing tool for slimming jawlines and altering facial appearance in images. With AI, he said, the possibilities are far more expansive because images can be regenerated rather than merely edited.

The social consequence, in his account, is that the edited image can become detached from reality. He said he had heard of someone meeting a person from Instagram whose photos looked nothing like her in real life.

Chris Williamson extended the point from individual editing to group dynamics. Citing writer Freya India, who had recently released a book called Girls, he said groups of young women taking party photos may fight over whose phone is used, because the phone owner controls the Facetuning.

Mack supplied the obvious incentive: the person with control can work on themselves “a little bit more.” Nirav Savjani added that it is now a significant social faux pas to post a picture in which you look good and someone else looks bad.

Williamson and Mack used Ashley Cole’s awkward AS Roma squad photo as a pre-AI example of how much social meaning can attach to one image. The screenshot shown on screen was a Bleacher Report article headlined, “Ashley Cole Appears to Be an Outcast in Incredibly Awkward Roma Team Picture,” with Cole standing apart on the far-left edge of the team photo.

Mack said the image became a major meme because the rest of the team looked close while Cole appeared awkwardly outside the group. He added that when he and Williamson are out with groups of men, he sometimes does “the Ashley Cole” by standing a couple of yards away and leaning in, ruining the photo.

The example was comic, but it clarified the pressure already attached to images. A single photo can carry an unwanted social meaning: outsider, awkward presence, bad friend, person who posted the unflattering version. Savjani’s concern was that AI and photo-editing workflows make correction easier and more expected; his example was that they can also make the resulting image less representative of the person.

George Mack connected his original meditation practice to a different problem: how machines should interpret signals they receive from a person. That question sits near the face-and-photo discussion because both depend on a machine reading an external signal — a skin image, a face, a photograph, or eventually brain activity — and turning it into a recommendation or representation.

Meditation had begun for Mack as part of the attempt to manage the skin condition. What remained was a question about self-observation: ask the mind what thought will come next. He described a brief silence — “six seconds” — followed by an apparently random mental image of former Bayern Munich winger Arjen Robben cutting inside onto his left foot. The point was authorship. If a brain-computer interface such as Neuralink can read or transmit thoughts, Mack wondered, how would it distinguish intentional thought from “monkey mind” noise?

Tim Ferriss turned that into a joke about failed public demos: someone gets on stage and the stream becomes an involuntary loop of obscenities. Williamson added a cruder version, and Ferriss joked that the likelihood of Elon Musk doing the first presentation made the risk more vivid.

Mack’s example from I Swear, a British film about a man with severe Tourette’s, kept the same question in view. He described a man growing up in Scotland in the 1970s and 1980s, when people around him did not understand the condition. In Mack’s account, the man walks to the shops, involuntarily insults people, tries to correct himself, and gets beaten up because he cannot explain what is happening. In another scene, while walking a beloved dog near a busy road, the Tourette’s produces a command for the dog to walk forward; he has to physically stop it.

Williamson summarized the conflict: “He’s fighting against himself.” That was the sharper version of Mack’s Neuralink question. A signal can be real without being intended. Mack’s concern was not whether a machine could detect activity, but whether it could tell the difference between what a person means and what the mind or body produces without consent.