AI hardware and software vendor Nvidia Systems has introduced new NIM microservices to the Universal Scene Description, or OpenUSD, standard for metaverse visual applications.
Nvidia on July 29, during SIGGRAPH, the computer graphics conference, revealed that generative AI models for OpenUSD development will be available as Nvidia NIM microservices. They are in preview now.
The move comes after Nvidia introduced the microservices at its GTC developer conference earlier this year.
NIM Microservices for USD
NIM microservices enable enterprises to create and deploy custom applications on their platforms.
The new OpenUSD NIMs will let to incorporate generative AI copilots and agents into USD workflows.
Microservices include USD code NIM, USD Search NIM and USD Validate NIM, all available in preview.
USD Code NIM microservice answers general USD questions and generates OpenUSD Python code based on text prompts.
USD Search NIM microservice lets developers search through massive libraries of OpenUSD and image data using natural language or image inputs.
USD Validate NIM microservice checks whether files uploaded are compatible with USD release versions.
Other microservices such as USD Layout NIM, USD SmartMaterial NIM and fvDB Mesh Generation NIM will be available soon.
Targeting the metaverse
Unlike the generative AI boom, the metaverse failed to gain immediate wide popularity, and remains largely confined to video headsets for virtual and augmented reality and some industrial applications such as digital twins.
In that context, the expansion of NIM microservice shows both Nvidia's commitment to generative AI and its ambitions in the physical and digital world, said Forrester analyst Charlie Dai.
"For the metaverse, Nvidia’s Omniverse platform continues to be a cornerstone of their strategy to enable the creation and connection of 3D virtual worlds," Dai said. "These microservices are one of the steppingstones on this journey."
One challenge for the metaverse is there is the lack of standardization to bring together the elastic, scalable infrastructure, compute power, storage, and data of the virtual environment.
This made USD for 3D and metaverse data interchange formats difficult, according to Constellation Research analyst Andy Thurai.
So, with its NIM microservice, "Nvidia hopes to bring generative AI capabilities to the robotics, metaverse, industrial design, and digital twin capabilities," Thurai said.
With the visualization and simulation of environments through the USD Code NIM microservice, Nvidia can help users revisit parts of the metaverse that were too difficult to develop before such as the virtual and augmented reality worlds, Thurai added.
However, adoption will be the biggest challenge for the AI vendor.
"The industrial areas they are taking on are too many and are very distributed both in the technology and in standards," Thurai said. "It is going to be extremely difficult to convince [customers] to adopt this."
Meanwhile, the Alliance for Open USD was created to help industrial companies adopt advanced technologies like the metaverse, he added.
Other than supporting the industrial metaverse, Nvidia is also looking ahead, Thurai said.
Generative AI appears to be slowing down in the adoption phase, and enterprises are not adopting the technology at the same pace they were experimenting with it, Thurai continued.
"If the market slows down, it could hit Nvidia hard," he said. "They are staying ahead of the curve by thinking and innovating this and being a market maker again."
Partner news
In another development, Nvidia's partner Getty Images also revealed on July 29 that it updated its generative AI image-generating model.
The updated model is built on the Nvidia Edify model architecture. Edify is part of Nvidia Picasso, a platform for building and deploying generative AI models for visual design.
Generative AI by Getty Images and Generative AI by iStock, also from Getty Images, are now updated with image-generating speeds of about six seconds, enhanced detail in generated images, longer prompt support and more control over output using shot type and depth of field.
Users can also modify both generated AI images and existing pre-shot images.
Nvidia also introduced fVDB, a deep learning framework for generating AI-ready virtual representations of the real world.
The AI vendor also revealed that Hugging Face will offer Developers Inference-as-a-Service Powered by Nvidia NIM.