Business is booming.

A Man Made an AI-Generated Design Firm in One Weekend — Take a Look


  • Alex Cornell, an interface designer, created a fake design firm with AI tools in one weekend.
  • Cornell said he used OpenAI’s GPT-4 and AI art tool, Midjourney for the project.
  • Some Twitter users thought the firm was real despite the images’ basic flaws.

Alex Cornell, an interface designer, made a fictional design firm in one weekend with the help of generative AI tools.

The design firm, Andever Design Partners, was so convincing some Twitter users assumed it was real.

AI-generated kitchen

A kitchen in one of Alex Cornell’s AI-generated houses.

Alex Cornell / Andever



“Some people thought it was a real business,” Cornell told Insider.”They were asking me how I was going to get clients and what happens when I need to buy furniture.”

Cornell, who used Midjourney and ChatGPT for the project, said it was a way to try and understand the tools for future work projects. 

“I think some people misunderstood my intent,” he said. “I got really good really fast using it just because I had such a specific target, it made it really easy to learn how to improve.”

He said the whole project took a weekend which “just speaks to how easy it is.”

“The dominant opinion was basically some kind of surprise that these tools are as good as they are. No one’s impressed with me, we’re all sort of collectively impressed with the tools,” he said. 

The design firm looked real 

Cornell said Midjourney, a generative AI art tool, was so easy to use it didn’t require any training.

“You go online and you’ll see a lot of people with video tutorials and tweet threads and all these things telling you how to get good at prompting,” he said. “It’s clear very quickly that all of those tutorials are useless — it’s so easy to use that you could just write cool bathroom and you’d have a viable image.”

AI-generated bathroom

A bathroom created with AI-art tool, Midjourney.

Alex Cornell / Andever



However, Cornell did discover a few tricks to keep things consistent. 

“If there’s any monochrome of skill involved, it’s figuring out how to create consistent residents,” he said, adding that using the same adjectives in prompts helped.

“Minimal was an important one,” he said. “Concrete was one I would use often.”

AI-generated bedroom

A bedroom created with AI tools.

Alex Cornell / Andever



He said it was important to have the same prompt for the background so the views out of the windows stayed the same.

“For the Lake Tahoe house, as long as every room sort of has some perspective on a lake most people wouldn’t really pick it up,” he said. 

“I found it very difficult to do a specific urban city out the window, for example, San Francisco apartment is difficult but if you just prompt city apartment you can’t tell where it is,” he said.

Cornell also created a website for the design firm. He used OpenAI’s GPT—4 to write the company’s history, create staff biographies, and suggest some prompts for Midjourney. 

AI still has basic issues

AI image generators have basic flaws.

The tools have infamously struggled to replicate human hands and can sometimes produce images with a plasticky look. 

Cornell said when generating interiors, the issues are slightly different.

He said: “You start to realize you probably shouldn’t have lighting or fixtures of any kind like faucets — you’ll see they are really contorted if you look closely.”

AI-generated kitchen

“Lighting or fixtures of any kind like faucets — they are really contorted if you look closely,” Cornell said.

Alex Cornell / Andever



“Legs of chairs and table legs, if you look closely, they often kind of blend together so the leg of the table is also the leg of a chair.”

He also said the bot was much better at producing bathrooms if prompted with the phrase “powder room.”

AI-generated bathroom

Cornell said it was hard to correct the small flaws.

Alex Cornell / Andever



“Sometimes it would put the toilet next to the bathtub. It was all the right ingredients but not properly positioned,” he said.

Cornell said it was frustratingly difficult to correct the small flaws, despite the overall images being impressive. 

“It’s trying to capture this whole image,” he said, “and it’s just kind of losing it a bit on the details.”





Source link

Comments are closed, but trackbacks and pingbacks are open.