Nvidia is offering businesses cloud services to develop generative AI models with their own data. Announced at the company’s annual GPU Technology Conference, the services are aimed at tasks such as writing ad copy, answering customer questions and generating images, video and 3D objects.
The two services, NeMo and Picasso, package pretrained models and development tools with access to Nvidia engineering support. Businesses can use Nvidia’s cloud to build and run models, then deploy them in Nvidia’s cloud or elsewhere.
Two services for different kinds of generation
Nvidia AI Foundations brings together NeMo, for language models, and Picasso, for generating image, video and 3D content. Both services include frameworks for processing data, APIs and pretrained models. That combination is intended to help companies adapt models to particular needs without building every piece of the system themselves.
NeMo is designed for text-based applications. Nvidia says developers can add specialized knowledge and functional skills to language models for chatbots, enterprise search and customer service. The company also says the models will be updated regularly with additional training data and can use information retrieval to draw on real-time data.
Picasso focuses on visual and 3D output. Customers can train and deploy models using their own data, starting either from Nvidia’s pretrained models or from models they already have and want to optimize. The service supports text-to-image, text-to-video and text-to-3D object models.
Access is limited as the services roll out
At the time of the announcement, NeMo was in early access and Picasso was in private preview. Businesses interested in either service had to submit an application, so neither was described as broadly available.
The rollout gives Nvidia a way to offer more than model access alone: the services combine pretrained systems with data-processing frameworks, APIs and engineering assistance. For companies, that package could support adapting generative AI to specialized work. The source describes the services’ capabilities and intended uses, but does not give prices or explain how applications are selected.
Creative industry partnerships emphasize content rights
Nvidia also said it was working with Adobe to co-develop commercially viable generative AI models on Picasso. According to Nvidia, the models would reach customers through Photoshop, Premiere Pro and After Effects, adding to Adobe’s recently debuted Firefly family of generative AI products.
Another partnership addresses how training content is sourced and how creators may be compensated. Nvidia said it was collaborating with Getty Images on text-to-image and text-to-video models trained using material from Getty’s library. Getty Images would provide royalties to artists on revenue generated by the models.
The arrangement comes as Getty Images is suing Stability AI over the use of Getty images to train Stable Diffusion without permission from Getty or rightsholders. Nvidia’s announcement presents a different approach: build the models with a content library whose relationship to the project includes artist royalties.
Shutterstock plans a text-to-3D model
Nvidia also plans to work with Shutterstock on a generative text-to-3D object model using Picasso. Shutterstock’s assets and metadata will be used for training, and Nvidia said Shutterstock would compensate artists through its recently launched Contributor Fund.
Shutterstock intends to offer the model on its own platform when it is ready for deployment. The company already provides a text-to-image tool through a partnership and licensing agreement with OpenAI, so the planned 3D model would add another type of generative output to its platform.
Together, the announcements position AI Foundations as both a model-building service and a platform for industry-specific projects. NeMo targets language tasks, while Picasso supports visual and 3D generation. The partnerships show how Nvidia and its collaborators plan to connect these tools to existing creative software and training content, with licensing and artist compensation forming part of the announced arrangements.