Now, somewhere in a faraway data center, a computer is depicting a room. Using cutting-edge machine learning techniques, algorithmic wizardry, and computer vision, we delve into the contents of a modern farmhouse kitchen and catalog its elements. tables, chairs, subway tiles.
This process is part of an artificial intelligence training program designed by a startup called Reimagine Home. Their hope is that by teaching AI to understand the contents of a room in natural language, it will better understand what humans really mean when they say, “I want a modern farmhouse kitchen.” That’s it.
Trippy, right? However, these are uncertain times. Last year, artificial intelligence engines such as GPT-3 and Stable Diffusion moved from research labs to the general public, rapidly beginning their transformation into consumer products. Some, like his Lensa, an AI-powered selfie adjustment feature, have become viral hits on social media. Some people write marketing copy or fight parking tickets. Of course, there are also people who do interior design.
The first such platform to attract public attention was Interior AI, which debuted in October last year. At the time, its designer and developer Peter Revels said: family business That his creation is only the beginning. ”[Image-generating AI models] Stable Diffusion and Dall-E are just a few weeks or months old. It’s very fast. We envision more products in the near future. ”
The level was correct. A mini-boom for AI-powered interior design apps has already begun. Some companies, like Reimagine Home and CollovGPT, were born from existing companies. Some are built by independent developers, such as RoomGPT. There are serious endeavors for founders who are giving it their all, and there are smart side projects for entrepreneurs who are still working their day jobs.
No matter the size or ambition, there’s no doubt that these apps have an audience. Users are flocking to her AI design site. According to founder Hassan El Mughalli, RoomGPT’s site has been visited by 3 million people, and founder Akhilesh Majumdar says Reimagine Home has been around for less than three months. However, it reportedly has hundreds of thousands of active users. Collov spokesperson Markk Tong said the company’s AI-designed chatbot was so popular that it was temporarily shut down for fear of overloading the site.
Much of that attention has been driven by viral social media posts and cultural hype surrounding artificial intelligence, but much of that has likely died down and these are the driving forces behind proving its longevity to consumers. It will be left on the platform. But professional viewers watch it as well. The founders of these startups have all seen their sites flock to real estate agents who constantly need to list properties virtually. Furniture retailers looking for ways to insert their products into AI-generated images are also interested. RoomGPT founder Hassan El Mugali said several brands have already inquired about collaboration.
Interior designers also use these sites in interesting ways. “We were contacted by dozens of designers,” says El Mugari. “Some of them have a lot of demand and need to decide which potential customers to spend their time on. But now they can use RoomGPT to create quick mockups. , we can now ask clients if they are interested, start engaging, and understand how serious they are.”
Money can be made in this field. Currently, most AI design platforms offer new users a few free mockups and then charge a few dollars in subscriptions or credits, which can add up to tens of thousands of dollars in the long run. Beyond that, looming are potential investments from a swarm of venture capitalists who have abandoned cryptocurrencies and are now going all-in on AI. A venture capitalist offered El Mghari $1 million at his $5 million valuation. This is a huge return on investment considering the fact that he built his RoomGPT in his spare time.
For founders, the promise of AI-powered “killer apps” (tech industry parlance for tools so powerful that they become ubiquitous) offers nearly unlimited returns. “The biggest potential impact I feel is to empower the average person who thinks they’re a great designer but isn’t. That’s exactly what I used to be. It’s a waste of human effort and capital,” says Majumdar. “If you can give that person a no-brainer tool that costs him maybe $200 instead of $2,000, that’s a huge monetizable market.”
Generated by Spacejoy
Generated by RoomGPT
left: Generated by Spacejoy | Right: Generated by RoomGPT
So what exactly can these programs do? Each has its own twist, but the basics are simple. A user uploads an image of a room and enters a few prompts (such as “modern living room” or “minimalist kitchen”), and the algorithm generates a rendering of the design for her. An early version of that concept was released with interior AI last October, but the technology is rapidly advancing.
First, AI is already much better at contracts. The debut version of Interior AI tends to rearrange room structures (windows, doors, ceiling heights, etc.) to achieve the desired look, creating an image that is stylistically accurate but not functional. Left by user.
“[The earliest apps] It just wasn’t very good,” El Mugari said. “The results were terrible. The generated room didn’t even look like the original room.” Since then, with the debut of a protocol called ControlNet, developers have Now we can put some kind of “strings” on how one, Stable Diffusion, works. Now, his AI design tools like El Mghari’s RoomGPT are very good at closely following the original parameters of a space.
Majumdar’s Reimagine Home takes that control a step further by automatically detecting furniture in a space and replacing specific pieces, rather than redoing the entire room. “Useful use requires precise control,” Majumdar says, noting that most users want to keep at least some of their existing décor.
ColovGPT is perhaps the most ambitious AI-powered design tool on the market today. Like anything else, you can detect the basic outline of a room and redesign it in different styles. However, the developers have rolled out various additional features. One of his is product detection. This is an AI-generated tool that identifies items in a room and generates links to real-world analogs.
The other is a chatbot that aims to integrate ChatGPT into the design process.In the demo of baud, the bot was able to generate a convincing “modern living room” and respond to basic requests like “remove the sofa” or “paint the walls red” with relative accuracy. It was far from perfect. CollovGPT couldn’t handle more subtle requests and was prone to making strange mistakes (for example, it made the walls of the room red, as well as the art and the sofa). However, this feature provided an interesting glimpse into a future where users can workshop design challenges in real time using his AI.
Generated by Spacejoy
Generated by Reimagine Home
left: Generated by Spacejoy | Right: Generated by Reimagine Home
AI-powered design apps bring cutting-edge technology to life. It feels like an incredible technological leap forward. It also has a lot of bugs and limitations.
Some of the weaknesses in these programs have nothing to do with artificial intelligence. All are more than a few months old, and the sites are riddled with routine glitches like pages not loading, images not downloading, and clunky interfaces. This is very much beta software, built on the fly and shown frequently.
The AI components of these sites also have many limitations. Generally, the more ambitious the concept, the more room for error there is. For example, CollovGPT’s “Product Identification” tool is a great idea in theory. The AI generates a chandelier, and when you click on it, you’ll see shoppable links to similar products. But in reality, this tool doesn’t really work. They have a hard time clearly identifying specific products, and tend to think of sofas as chairs, for example, much less recommending similar products. (Tong, the Collov spokesperson, acknowledged the tool’s limitations and said that, like other AI technologies, it is a work in progress, adding that “we will continue to improve it in the coming months.”) We are definitely moving towards a new version.”)
All of these apps also suffer from a lack of differentiation. While each has their own unique spin, they all utilize the same AI engine to power their work, giving them a level of sameness. “We’re all doing the same thing. We’re all using his ControlNet,” he says. “There’s also an element of just waiting.” [new AI technology] come out. “
Collov is working to overcome this problem by hosting “design hackathons” that invite users to upload design images and tag them by style. Ostensibly a contest with a $1,000 prize, the effort’s purpose is clearly to train the company’s AI. . El Mghari is committed to creating better tools for designers. Meanwhile, Reimagine Home is working on customizing its models and is experimenting with having AI write room descriptions.
Over time, developers hope that efforts like this will make their creations stand out from the rest. “AI is a foundational technology and something you build on top of,” Majumdar said, adding that core technologies such as operating systems and batteries will serve as the foundation for an entire ecosystem of specialized products. I’m using an example. “It takes a lot of effort to truly understand what customers need and solve them. That’s a huge opportunity.”
But despite his enthusiasm for the possibilities, Majumdar is also candid about the challenges. The race to build AI-powered design engines is heating up, but the finish line is never in sight. “Right now, it’s a great toy,” he says. “But there’s still a chasm to cross before it becomes really useful.
Homepage photo: Created by Midjourney