Researchers showcased AI technology that enables fast creation and editing of materials, accelerating 3D workflows. At SIGGRAPH, NVIDIA's researchers presented a generative AI workflow that assists artists in swiftly creating and iteratively processing 3D scene materials.

This research demonstration took place at the Real-Time Live event at SIGGRAPH, showcasing how artists can use text or image cues to expedite the generation of customized texture materials such as fabrics, woods, and stones, while maintaining finer creative control. NVIDIA Picasso will offer these capabilities, enabling enterprises, software developers, and service providers to utilize fully licensed data in developing customized generative AI material models.

This AI model will facilitate iterative processes in material creation and editing, empowering enterprises to provide new tools to aid artists in swiftly refining the appearance of 3D objects until achieving the desired outcome.

During the demonstration, NVIDIA researchers, assisted by AI, crafted a living room scene resembling an interior designer's work using a 3D rendering application. The researchers utilized NVIDIA Omniverse USD Composer—a reference application using Universal Scene Description (USD)—to assemble and structure the scene. They employed this application to add brick-textured walls, create and modify fabric options for sofas and cushions, and introduce an abstract animal design in a specific wall area.

Generative AI Driving Iterative Design

The Real-Time Live demonstration combined several optimized AI models. Developers will be able to customize these tools using Picasso and integrate them into creative applications tailored for artists. Once integrated, artists only need to input brief text cues to generate materials like bricks or mosaic patterns. These materials can be tiled seamlessly across surfaces of any size. They can also import reference images, such as a piece of velvet fabric, and apply it to any object within a virtual scene.

AI editing tools enable artists to modify specific areas of the materials they're working with, such as the center of a coffee table texture. These AI-generated materials support physically based rendering, realistically responding to variations in scene lighting. Features like normals, roughness, and environment occlusion maps are crucial for creating and fine-tuning realistic 3D scene materials.

Accelerated by NVIDIA Tensor Core GPUs, artists can generate materials almost instantly and upgrade them in the background, achieving resolutions up to 4K, while creators continue perfecting other parts of the scene.

In creative industries like architecture, game development, and interior design, these capabilities aid artists in swiftly exploring creativity and experimenting with various aesthetic styles, creating multiple versions of scenes. For instance, game developers can expedite open-world environment design or create character costumes using these generative AI features. Architects can experiment with different architectural facade designs in various lighting environments.

Building Generative AI Services with NVIDIA Picasso

These physically based material generation functionalities will be available in NVIDIA Picasso. Picasso is a cloud platform where enterprises can build, optimize, and fine-tune generative AI foundational models for visual content. Using Picasso, content providers can develop and train generative AI tools and services based on fully licensed proprietary data. Picasso is part of NVIDIA AI Foundations, a suite of model-making services aimed at advancing the application of generative AI in text, visual content, and biological fields.

During the SIGGRAPH keynote address, NVIDIA CEO Jensen Huang also announced a new Picasso feature—generating lifelike 360 HDRi environment maps based on simple text or image prompts to enhance 3D scenes.

*This article is reproduced from NVIDIA Enterprise Solutions.

*Copyrights for images or videos related to NVIDIA products (in whole or in part) belong to NVIDIA Corporation.