Pay-per-use physics models for virtual fashion
19.46, Wednesday 21 Oct 2020 Link to this post
How about skin tight t-shirts with tracking markers, especially made for rendering synthetic shirts with physics-model fabric for wearing on Zoom? And how would the virtual shirts be priced?
Problem. I spend my work days on video calls. For some of those calls, I wear a shirt. I want to give a good impression, and a shirt is good for that, but it feels like a waste: I have a limited number of shirts before I have to take them off to the cleaner. (Getting my shirts cleaned and pressed is my laundry vice. Generally laundry is my happy place: sorting, folding, all that, it’s a finite task that I love. But ironing bores me to tears, so I factor into any shirt purchase the fact that I’ll have to pay for cleaning.)
Solution. Virtual fashion.
The concept is that I could wear a skintight t-shirt printed with computer vision motion tracking markers, semi-reflective patches (to measure ambient light), and known colour areas (for white balance).
Software on my computer would intercept the webcam signal, and add a virtual shirt – or a virtual anything else. Using a physics engine, it would have a full cloth simulation to mimic light or heavy fabrics (the shirt I wear depends on the season), and adapt to my movement and the light in the room.
The idea is that, for everyone else on the call, the virtual shirt is indistinguishable from me wearing an actual shirt. Only I have an infinite wardrobe.
Shirts would be purchased from all the usual designers and retailers: Hugo Boss or Uniqlo, whatever. Buy as normal, but download the 3D model into your virtual fashion software, and there’s no physical garment to take to the cleaners.
This concept at least partially inspired by a recently-purchased all-over-print ugly shirt:
a ridiculous-looking garment that magically renders the wearer invisible to CCTV.
Here’s a pic of me in the t-shirt. Ostensibly the pattern confuses the facial recognition algorithms of a certain brand of CCTV camera. This article talks more about the patterns and links to the original paper.
So if there are patterns that computers see badly, are there patterns that computers see really, really well, and what would you do with that? Hence a motion cap shirt for virtual fashion.
The question is: how would you price a virtual shirt? Is it a one-off purchase, or perhaps a subscription to a virtual wardrobe?
For me, the key difference with garments (over, say, music) is that’s it’s possible to spend more on the material itself and that expense is visible to the trained eye.
Sure, expensive material can mean it hangs better, or lasts longer, or whatever. But high fashion doesn’t always do those things. And sometimes exactly what you want is a low-cost basic: higher price doesn’t always mean better.
So what unfakeably expensive material provides, if nothing else is a meaningful foundation for wide price differences, and that gives rise to exclusivity, brands, and all the rest.
How to replicate all of that with virtual fashion?
With virtual garments, there’s no meaningful reason to price a Prada shirt differently from a no-brand one. The design maybe? But the value of design is down to personal preference; there will be no consensus on what should cost more and what should cost less.
UNLESS: The true difference between virtual garments is down to the quality of the simulation.
A simulation with more compute thrown at it can and will look better. Throw a better GPU (the chip responsible for rendering the graphics) at the rendering problem, and the virtual shirt will be higher resolution, run at a higher frame rate, and the cloth will hang more authentically. It gets rendered once, on your machine, and then everyone else on the video calls gets to see it. And better GPUs do indeed cost more. Nobody else needs an expensive CPU, but they’ll definitely be able to tell that you spend a lot on yours.
The difficulty is that the GPU is bundled with your laptop or smartphone. Any virtual garment, high end or low end, gets rendered with the same chip. Quality difference is eliminated.
Pay-per-use physics simulations
So how about this, to open up the economics of virtual fashion:
Speculatively bundle the absolutely best available GPU in every laptop and every phone. But don’t activate its full capabilities, and don’t pass the cost onto the consumer at the time of purchase. Instead allow virtual garments (which are just 3D models in physics simulations, don’t forget) to pay to unlock levels of capability on a per-model basis.
The 3D model designer would pay the GPU manufacturer directly. Perhaps they pay a per-unit fee with a multiple that takes into account model complexity and level of desired verisimilitude.
The consumer, purchasing the 3D model, would pay the designer. When they use the item, the GPU is unlocked for that item only. The consumer would not be able to purchase a garment and then choose how much rendering power to give it.
Instead of an upfront GPU purchase, this is a pay-per-use model.
It would allow for purchasing cheap virtual shirts that look ok, and expensive virtual shirts that look shimmering, amazing, and computationally profligate.
(What do we call this? Speculative economics?)
I wonder what high fashion would look like in this world?
Anything that requires top-end GPU capabilities I imagine: lots of reflections, lots of crinkles and complexity on the surface, and lots of semi transparent layers, all falling in interesting (and expensively) crumpled folds.
And then I wonder whether real-life fashion would end up mimicking the virtual? Would high-end garments end up using ordinarily-priced material… but all designed to appear especially difficult to render in simulation?
All of which is to say: I would pay money for an actual t-shirt which is designed to be hard to visually simulate. Do let me know if you make one.