This is why Autodesk is obsessed with geometry and reflects this “obsession” in its software.
Today, Autodesk unveils a first-of-its-kind research project called “Bernini,” and its first experimental generative AI model that quickly generates superior 3D shapes from a variety of inputs. Including a single 2D image, multiple images showing different views of an object, point clouds, voxels and text. The first Bernini model is ideally suited for professional geometric workflows and can generate multiple functional variations of a 3D shape from a given input.
Autodesk is working on creating generative models that could be used for different use cases in architecture, product design, entertainment, and more. The company focuses on generating functional 3D structures because items built or manufactured from the results of these models must function in the real world and serve the purpose the designer has set.
A simple example is that of a pitcher of water. Many other 3D generative models can produce shapes that resemble a jug with textures that enhance their surface appearance in a very specific light environment. But the Bernini model generates shape and texture separately, without confusing or merging these variables. So the water pitchers generated by our model are hollow in the middle and can actually hold water, as any real-world pitcher should.
Just as importantly, Autodesk is focused on bringing a truly generative approach that produces multiple variations, providing choice to the designer and contributing to their creative workflow. According to 2024 State of Design & Make report from Autodesk, 78% of business leaders believe AI will improve their industry and 79% agree that AI will make their industry more creative.
The Autodesk AI Lab, part of Autodesk Research, is a multi-disciplinary team of scientists and specialists working on the application of AI to facilitate difficult tasks in design, manufacturing, architecture, engineering and construction, as well as creative processes in media and entertainment. The researchers trained the Bernini model on ten million different 3D shapes, a composite dataset made up of publicly available data, a mix of CAD objects and organic shapes. The research that contributed to this work was published earlier this year by the AI Lab in collaboration with the Chinese University of Hong Kong.
At this stage, the Bernini project is strictly experimental and is not available for public use.
As the team works to refine this technology, Autodesk is actively engaging potential partners to study and improve the performance of generative models for design and manufacturing. While the Bernini project is exciting and at the forefront of generative AI, its models will become increasingly useful and compelling when trained on larger and better professional datasets. The Bernini model is specifically structured to provide effective training and fine tuning to make this task easier.
If trained on buildings, the models could generate creative, geometrically rigorous designs and inspire a new generation of buildings and architects. If they are trained on video game character models or fantasy environments, they could produce fascinating new creatures or virtual worlds. If they are trained in car design, they could help imagine a new series of innovative vehicles.
Project Bernini is part of an overall Autodesk AI strategy to provide intelligent assistance and generative capabilities spanning the entire product portfolio. Throughout this work, Autodesk is committed to integrating safety, transparency, and humanity into the development and use of AI.