AniML, the French startup behind a new 3D capture app called Doly, wants to create a kind of PhotoRoom out of product videos. If you sell sneakers on an online marketplace or need to create direct-to-consumer Instagram ads for your merchandise, Doly helps you generate 3D models with your phone and turn them into professional-looking product videos.
While creating a video is extremely difficult, generating a 3D model is even more difficult. That’s why the AniML team focused on simplifying the experience. They want to make 3D capture a mainstream technology by starting by packaging it into an iPhone app.
Here’s how 3D capture with Doly works: The user points their phone’s camera at the product and physically moves around it to capture it in 3D. Behind the scenes, the app grabs still images and sends them to the cloud. AniML built a reconstruction pipeline using something called Gaussian splatting to turn these images into a realistic 3D model.
3D models are traditionally created with a collection of points in 3D space, some 2D texture projected onto those surfaces, and lighting effects. Gaussian splatting is an entirely new rendering pipeline that involves estimating a 3D point cloud from a set of 2D images using a pre-trained AI model.
“Our starting point was a technological breakthrough: AI had just arrived in the 3D world. So people at Facebook, but even more so at Google, did research and wrote a pretty important research paper on something called NeRF,” AniML co-founder and CEO Remy Russo told TechCrunch. “It’s a new paradigm where you try to reconstruct 3D by letting machine learning do the work.”
“You’re no longer working in polygon-based 3D, but now you’re in neural-based 3D,” he added.
Gaussian splatting isn’t exactly the same as NeRF, but it’s sort of a legacy 3D modeling technology, as Rousseau puts it.
This is the technical part. AniML then focused on finding a use case that could grab users from day one. E-commerce companies were the obvious choice for a 3D modeling tool.
What else does the app offer? After capturing a 3D model, Doly users can browse a template library to select a 3D scene to integrate their object into. This can be a simple 3D spin with a plain background or something more dramatic, from a marketing point of view, such as the camera slowly zooming in on the subject and switching to different angles.
If a customer likes the result, they have the option to buy the video from the app and download it for use elsewhere.
Russo previously founded two VR companies – including Mimesys, a startup that was acquired by Magic Leap in 2019. His co-founder Pierre Pontevia also has an interesting background, having sold a company to 3D tools giant Autodesk; and another to the 3D content development platform, Unity.
AniML has raised $2 million so far, with Adjacent leading the seed round. The startup also participates in AI Grant, the startup accelerator led by Nat Friedman and Daniel Gross. Also investing are Kima Ventures and several angel investors, including Hugging Face’s Julien Chaumond; Nicolas Steegman and François Lagunas, who previously founded Stupeflix; Alban Denoyel of Sketchfab fame; Bertrand Schmidt; Thibault Elsier; and Vincent Nalatambi. We are also told that Bpifrance has contributed part of this grant round.
It will be interesting to see if big brands, second-hand distributors and other e-commerce professionals adopt 3D rendered videos for upcoming campaigns and online listings. But now it’s nice to see that you may not need a professional video studio to create compelling product visuals thanks to artificial intelligence.
