Ever wanted to create 3D models without learning the complicated ins and outs of industry tools like Maya? NVIDIA is making it possible with Magic3D.
The new AI can take user input in the form of a text prompt and use it to generate a 3D mesh model. Usually, building a 3D mesh model is a tedious process involving tens, if not hundreds, of hours of work. However, Magic3D can create a complete mesh model with colored texture in only 40 minutes.
Artists will still need to modify these AI-generated models before implementing them in a video game or CGI scene, but they’ll still save an excessive amount of time.
Additionally, artists can use Magic 3D can edit 3D meshes by tweaking the text prompt for the model. The AI can also apply 2D images to a 3D model and preserve a model through multiple iterations. The team behind Magic3D believes that its AI will allow anyone to create 3D models in the future without any training.
“We hope with Magic3D, we can democratize 3D synthesis and open up everyone’s creativity in 3D content creation, ” Nvidia’s Magic3D team writes in an academic paper. According to said paper, Nvidia is developing Magic3D in response to Google’s DreamFusion, which takes text input and creates a 2D image which is then optimized via a neural radiance field (NeRF). Nvidia states that Magic3D can generate 3D models twice as fast as DreamFusion.
Nvidia’s Instant NeRF
Of course, Nvidia already has a NeRF-based tech that transforms 2D images into 3D scenes. The company combined NeRF technology with Inverse rendering (a process in which artificial intelligence recreates a 3D scene using 2D images taken from different angles) to create Instant NeRF. Instant NeRF requires just seconds to analyze 2D images to render a 3D scene.
But Nvidia isn’t the only company bringing 3D modeling to the general masses. CapturingReality’s Realityscan app allows users to snap photos with their smartphone to create high-fidelity 3D models.