If you’ve ever bought a sofa on an online store, have you thought about the homes that you can see in the background? When it’s time to release a new collection, furniture brands usually spend a small fortune on photo shoots. It’s a cumbersome and expensive process as it’s not easy to move furniture around.
That’s why a French startup called Presti is using generative AI to turn a single product image into a realistic photo. The company has already raised a $3.5 million seed round led by Partech with several business angels also participating.
“Very quickly, we picked up the phone and talked with 50 potential users,” Presti co-founder and CEO Nabil Toumi told TechCrunch. “And everyone was saying the same thing. Creating product visuals was a process that took a very long time, cost a lot of money and they didn’t have a simple solution for creating these photos. At the same time, it was really the most important asset for brands so that they can create a unique identity and sell online.”
Presti turns this:
Into this:
At first, Presti didn’t narrow down its focus to furniture companies. But the startup quickly realized that furniture companies faced some particularly difficult pain points.
“For a photoshoot, they needed to rent a nice house, they needed to transport products — so they had high logistics costs associated with that. And those photo shoots were planned months in advance and ended up costing them hundreds of thousands, even millions of euros a year,” Toumi said.
At its core, Presti uses Stable Diffusion XL as its foundation model. It has been retrained and tweaked so that it works particularly well for product imagery in the furniture industry.
At first, the team tried to use a vanilla version of Stable Diffusion XL. But they quickly realized that there were issues. “You’re going to have legs added to your sofa, and the backrest will be distorted,” Toumi said. Similarly, it was hard to get the perspective right. For instance, the wall behind the sofa needs to be parallel to the sofa.
“At the same time, something that is really important is the dataset on which we’ve trained our model. We currently have over 75,000 images of ultra-high quality photos of furniture in our industry, which we can use to train our model to reinforce the learning process for a specific use case, for this type of photos,” Toumi said.
Presti didn’t want to stop at background generation. Customers can also add accessories on their products. For instance, if you’re generating product shots for a new sofa, you can add cushions. These cushions will project a realistic shadow on the sofa so that they don’t look like something that was added in Photoshop.
Similarly, furniture brands usually have several variations of the same model with different textures and colors. While this is still a work in progress, Presti hopes that its customers will be able to swap the material using its tool. As a result, it’s going to be much easier to release new products for companies using Presti. Freelance photographers aren’t going to like this new product, though.
While Presti mostly works with mid-sized furniture companies, it also has a strategic partnership with Maisons du Monde, which is one of the largest furniture retailer in France. In addition to Partech, other investors in the startup include Thibaud Elzière, Maxime Brousse, Abou Laraki, Rémi Lemonnier and Julien Hirth.