A new experimental AI called GaussianDreamer promises to significantly speed up the creation of 3D models from text. In the concept known as “text-to-3D”, the user enters a text that describes the desired 3D model. The AI then generates a suitable 3D model from this.
Previous approaches often generated rough models and were limited to certain model types. According to the developers, GaussianDreamer combines two AI methods. First, a rough 3D model is generated. Then the new Gaussian Splatter technique refines the model.
With one GPU, GaussianDreamer can generate a high-quality 3D model in 25 minutes. The speed can be further increased by connecting several GPUs in parallel. According to the developers, GaussianDreamer paves the way for powerful text-to-3D systems that could be used by many. Easy access to 3D models is seen as the key to making 3D printing more attractive to consumers.
GaussianDreamer is still in the research phase. Interested parties can try out the AI via a public interface.