Fake food on the real menu
Fake food on the real menu
How artificial intelligence fills in the blanks
A restaurant tech company just introduced a new tool for restaurants using artificial intelligence (AI), the buzziest new technology you can’t avoid. It’s a way to generate images of food — any food — instantly, and it’s from Lunchbox, an online direct ordering platform known for flashy branding and stunt-y marketing efforts, like the real tattoo studio it ran inside its booth at last year’s National Restaurant Show.
Lunchbox didn’t build the technology, it instead created some restaurant-specific parameters that tap into popular software: DALL-E is a text-to-image generator from OpenAI, a company that exists to build AI models anyone (who pays) can use and understand. To use Lunchbox’s version, type in a few words: “chicken tenders / orange background” for example. You’ll quickly receive four different images of chicken tenders on an orange background (with varying degrees of edibility, as seen below).
In an interview, Lunchbox co-founder and CEO Nabeel Alamgir said the service is especially useful for quickly filling gaps on online menus or to promote specials and limited-time offers — online menus with food photos get more diner engagement, he noted. But while pulling an image from thin air fills a square on a webpage, it also creates an opportunity for disappointment after a restaurant sells a customer on a dish that doesn’t really exist.
Delivery giant DoorDash has its own data: in a study of over 15,000 small restaurants on the platform, menus with item photos saw up to 44 percent more monthly sales. In a 2021 blog post, the company spent nearly 1,000 words describing six food photography tips, including how to identify good lighting, the importance of showing the whole dish to communicate portion size, and, of course, a primer on editing and retouching, all in service of creating a compelling image that accurately represents a menu item. (There are other how-to guides of similar length all over the internet.)
This sudden focus on menu photography is just another way that online ordering and delivery apps have shaped the restaurant industry in the name of technological progress. Of course it’s true that listings with photos are more interesting to diners; the apps are literally designed around them in order to draw customers in with realistic photos of what to expect.
In fact, DoorDash explicitly disallows stock photography and clip art.
From its help pages: “Using graphics like clip art or stock animations of an item is not acceptable; we want customers to see exactly what they will receive.” Further, “Stock photos typically look great, but they are not a photo of your item specifically in addition to copyright issues, so they will be rejected.”
DoorDash competitor Uber Eats similarly encourages well-lit and representative photos — so much that for $125 it’ll send restaurants a vetted pro photographer for an hour-long session. Restaurants that do a brisk take-out business are right to incorporate food photography into their workflow; especially now that it can do double- or triple-duty on social media and other digital platforms. But it’s a time-intensive process. Is taking a tech-y shortcut so wrong?
I don’t know the answer, but the ethics go deeper than menu item fakeouts. Critics of OpenAI’s DALL-E image generator say it’s stealing intellectual property from artists and photographers. In July, as Engadget senior editor Daniel Cooper noted:
But there is a problem, since these systems are trained on existing material, often using content pulled from the internet, from us. Is it right, then, that the AIs of the future are able to produce something magical on the backs of our labor, potentially without our consent or compensation?
Yes, the images are reasonably representative and easy to obtain, but they’re based on real photos taken by real people of real food made by real people who were not compensated for contributing to the collective digital intelligence that could one day threaten their own artistic existence.
That might sound extreme, but even the easiest methods of taking photos of menu items require time, effort, money, and work. Lunchbox’s image-generating tool is free to use. OpenAI’s software, which does the heavy technological lifting, allows a set number of free image-generations per month; after that users can buy 115 more for $15. Lunchbox is covering the cost of its food images for users, “Because it’s expensive to get these API calls with OpenAI,” Alamgir said in that interview. According to OpenAI, one image large enough to use on a digital menu costs $0.02. As I mentioned, Lunchbox’s tool returned four options each time I searched. And to be fair, I did it twice.
This sort of AI, the type that makes you worry about how intelligent we’re making the machines, is in its infancy. The Lunchbox feature is a shortcut, a distraction, a fun toy. It’s clever, but not necessarily smart.
What else?
Grazzy, a payments startup that uses QR codes to encourage digital tipping for hourly workers, raised $4.25 million in a Seed round of funding. Grazzy makes money by charging a transaction fee; it’s sometimes paid by the customer, sometimes paid by the business, or, as a representative for the startup explained over email, the employee can be responsible for paying the fee. (Yikes.) But for now, “Grazzy is seeing that the hotels who are currently using the platform are handling the transaction fee so employees are taking home 100% of their tips,” the spokesperson wrote.
Chili’s closed its delivery-only store after 2 months. It was a test concept near a university in Texas, closed to “focus on innovation within our four walls,” per a company spokesperson.
New York’s City Council approved a bill limiting extras in to-go bags. The bill prohibits restaurants from including napkins, utensils, plates, and condiments with orders unless the customer requests them. (Anyone else nursing fond memories of well-stuffed New York bagels with 30 tiny napkins smashed inside a paper bag? Just me?)
Waffle House is not serving your off-menu TikTok hacks. Nope!