It was during one of my late-night perusals of the internet trying to stay on top of current Hermès scarf market trends, availability, and sales, that I came across a listing on eBay that stopped me in my tracks. It offered a wealth of information on how a 90 cm piece of silk could elevate my style and how, as a woman, I surely seek luxury, and so on. However, I noticed the definite absence of a detailed condition report or any of the valuable information that a serious buyer actually seeks.
I soon saw the exact same “description” verbatim in another listing, and then another. This was in 2023 and the result of eBay introducing AI as a seller tool, the “magic lister”, to write descriptions for sellers who did not want to bother writing their own. The result? A wordy filler lacking any pertinent information. It became a running joke with my husband; whenever I put on one of my Hermès scarves, he would say something like, “I see you are elevating your style today, my dear.” We still chuckle at it; somehow, the joke never gets old.
Almost three years later, I am struck by yet another AI phenomenon: the use of AI-generated photos. I call this dimensional dissonance, occurring when a seller lacks either a mannequin or a human model and wants to show the scarf being worn. In these cases, a photo of a 90 cm x 90 cm Hermès scarf is uploaded to an AI model, which is then asked to generate a photo of a person wearing it. The result is a “lovely” photo that is, unfortunately, not of the actual scarf, but merely a digital representation of it.
This is the image Gemini generated.
As I learned, generative models are taught on 2D models and they currently lack the ability to represent our 3D world accurately. As a result they struggle with scale because they treat the scarf as a texture rather than a physical object with fixed dimensions (such as 90 cm or 140 cm). This can lead to visual gaslighting, as the AI often produces an image where the scarf appears either far too large or far too small or has added design elements or lacks them. Consequently, details are lost, and text in particular often becomes unreadable.
Using one of my recently blogged about Hermès Sichuan carré photos, the results from both ChatGPT and Gemini when asked to “create photo with person wearing this scarf” can be found below.
And finally an ocher border was added to the original black border – why I ask, why?
When asked to scale the scarf in the image above down, ChatGPT provided this image … adding what can only be described as gibberish to the border…
The other things to note, and I realize as machine learning becomes more and more sophisticated, this might soon become a mute point, is the lack of clarity and detail in all of these examples. Above is a photo of an actual Decoupages scarf draped over my mannequin, Lucy, and below one generated by AI (Gemini). The writing in particular is illegible and the overall image of the scarf and details blurry.
Sadly, buyers who trust photos generated by AI, unless made aware of before hand, will, needless to say, be puzzled if not disappointed when they receive a scarf that is half or twice the size of what was pictured, a scarf that has additional or missing design elements thus creating unnecessary confusion and ultimately diminishing, and in my opinion, definitely not “elevating”, the credibility of the seller.
While reputable resellers (yours truly included) use professional photography to capture the beauty and details of an actual Hermès carré offered for sale, AI uses algorithms to create an illusion.
One is skill and expertise; the other a shortcut.
Thank you so much for your analysis of AI right now and the presentation of the H scarves. I feel like a more informed shopper and not just for Hermes scarves!
Thank you Cristina, for letting me know you found the article informative. With appreciation, Ladi