Introduction:
I'm constantly searching for tools and tech to make work/life easier so it was inevitable I would end up on the current bandwagon of choice. AI is everywhere at the moment so in order to learn more about the capabilities of the various available platforms out there, I came up with a test based on a personal task that tbh, I would fob off onto a virtual assistant if I had one. I'm talking of course about that great drain on the soul of every guy on the planet. The idea of getting home from a busy week at work and then at the weekend venturing out to the local trading mecca and spending several hours dodging people, bright-blinding sale posters, crowds of people who seem to want to collide with you and shops forcing their mind-altering suggestions of what you should be wearing onto your retinas rather than hiring knowledgeable advisors who can explain how a classic-fit is different to a regular fit. What should I be asking for though? I don't have time to become a style-guru so as encouraging people to part with their hard-earned is going to be the primary use of AI as far as all those internet marketers are concerned, why not bite the bullet now and see how bad things are going to get. Who knows, maybe we'll be pleasantly surprised.
So my experiment will ask three different AI platforms for some style guidance. I'll then get the platform to condense that into a prompt for Stable Diffusion online, so I can see what another AI thinks that looks like. After all, a picture is worth a thousand words. The platforms I've decided to put to the test are ChatGPT, Google Bard, and Perplexity.AI. Each platform will be rated based on the usefulness of the image although I will probably hit refresh a subjective number of times until I get something which I think is useful to me. I will provide each platform with the same information about me; size information, preferred colours etc. but the platforms will be tasked with figuring out what I should be wearing and I'll measure the success by the image produced from the output prompt.
So what are the common parameters. Well I'm around 5/7, not enviably slim and my favourite colours are light greys, browns, navy blues and blacks. Let's see how the platforms compare when it comes to smart-casual advice based on my frame. So, even though it would be best to tweak my prompts for each AI, I'm just going to go with this prompt and see what they come back with:
I'm a short guy around 5ft 7, not the slimmest frame but not obese. My favourite colors are light greys, browns, navy blue and black. Figure out what the best advice would be for a smart-casual outfit for me, with any accessories to complete the look. Turn this into a prompt for stable diffusion so I can see how this outfit would look on a shop mannequin
ChatGPT
ChatGPT is probably the most famous AI at this moment in time. Is it any good for clothing advice though since it was trained by scraping internet data up to 2021. This suggests it might not be the best choice for up-to-the-minute fashion advice. Then again classics never go out of fashion or so they say. I had to slightly tailor the prompt as there were just too many dresses and high-heeled shoes for my liking. So I ended up with this:
"Create a smart-casual outfit that perfectly suits a short guy with a height of around 5ft 7. The person has a moderate build and favours colors like light greys, browns, navy blue, and black. The outfit should be accompanied by suitable accessories to complete the overall look. Visualize this ensemble on a shop mannequin, allowing us to see how it all comes together in a stylish and cohesive manner."
Here's what Stable Diffusion decided that ChatGPT wanted to show me:
Ok so there's still high-heels in there but I could have put that in as a negative prompt. I think it's more important for the test to try to stick to the original output prompt as much as possible, but to be fair I opted to do the same tweaks with output prompts for the other platforms too if necessary. I think this was caused by ChatGPT deciding that I was an 'individual' rather than a guy. The lesson here may be that bias should not overrule fact and we should remember that Darwin ultimately died for the right to observe species and classify according to observations.
Google Bard
Google Bard was an interesting test. Before spewing out it's stable prompt, it decided to actually feed me some fashion advice and highlight a few products for me. In fact it suggested a pair of shoes which were very similar to a pair I bought in the last month. Here's the prompt which Bard decided to give me for Stable Diffusion:
"Create a 3D rendering of a shop mannequin wearing a light grey button-up shirt, dark brown chinos or jeans, brown leather loafers or boots, a brown leather belt, a pocket square, and a watch. The mannequin should be 5ft 7 and have a medium build. The colors should be light grey, dark brown, navy blue, and black."
And here is the result. I'm not sure about the 3D rendering as I actually thought the options Bard decided to show me before this stage were better than the rendered guesses. Maybe that show that an AI which is able to call up exact images should be judged on the original content rather than the alternate AI imagined image.
Perplexity.AI
Like Bard, Perplexity has access to the internet so in theory options could be from more recent sources. Like Bard, it also offered me some fashion tips before the prompt which were well presented and in bullet-point. I liked this as keeping the advice to short tips is really good for imprinting on your mind the characteristics of what suits you best. Here's the prompt Perplexity offered:
"Create a smart-casual outfit for a short guy around 5ft 7 with a not-too-slim frame. The outfit should include dark denim jeans, an unstructured navy blazer, a white shirt, and dark leather shoes in brown or black. Accessories should be kept to a minimum, with a leather belt in black or dark brown, a square in light or pale blue, a smart watch, and smart sunglasses as options."
Here's how that looks according to Stable Diffusion.
Conclusions:
I guess you have to keep in mind that these things are still very new to most of us. Based on these initial images though I think Perplexity currently comes out on top. ChatGPT's reluctance to accept I might be male even though I identified as 'a guy' no doubt threw a spanner in the works when it came to shoes. Bard's actual advice was great however it's attempt to convey that via a Stable Diffusion prompt was lacking. It's very early days for Bard though so let's wait and see what happens. Perplexity definitely came out on top for me with this test. Those images look like they could have come straight from 'Short-guy, fashion advice weekly' or some equivalent publication.
Resources:
