Luma AI’s Genie can generate 3D content from text prompts and is now available as a free research preview via Discord.
Silicon Valley startup Luma AI specializes in combining AI technologies with 3D content, such as generating NeRFs or other 3D formats from smartphone videos. Now the company has released a Discord bot called Genie for text-to-3D generation.
Genie allows users to enter a text command and within about 20 seconds receive four initial, simple 3D models, each with an interactive link to view or download as a 3D object in glb format. In addition, users have the option to further refine the model for improved resolution and more accurate results. This process currently takes around 20 minutes, according to Luma AI.
Text-to-3D: A revolution for game development and virtual production?
In the long term, Genie is expected to find applications in game development and virtual production. Until now, creating an extensive game world with complex 3D assets has been a time-consuming and demanding process, often requiring a team of specialists. With Genie, game developers can easily put their vision into words, the company says. This would allow artists to focus on improving assets rather than creating them from scratch.
Other applications Luma AI sees include virtual production studios. Currently, Genie is a “research preview” that the company will continue to improve, as well as the underlying foundation model, before launching it as a paid service.
Generative AI for 3D next target for OpenAI & Co.
Now that generative AI for text and images has transformed many industries and fields of work, generative models for 3D and video are seen as the next hurdle that research labs and companies want to overcome.
Text-to-3D models in particular, much like 2D image models such as Stable Diffusion, Midjourney, or DALL-E 3 before them, could primarily affect artists. The development of such models has been progressing steadily since at least 2021, with examples such as Google’s Dream Fields and Dreamfusion, CLIP-Mesh, OpenAI’s Point-E and Shap-E, Nvidia’s Magic3D, Tencent’s Dream3D or, most recently, 3D Gaussian Splatting.
Luma AI’s Genie is available via Discord. The link is available on Luma AI’s website.