AWS + NVIDIA Partner Innovator Story

This case study first appeared in AWS' Partner Innovator Stories. To see more Bria Case Studies, explore our repository here.

Balancing Speed, Scale, and Compliance

BRIA AI sought to train an AI model on licensed data to help customers produce creative content faster without compromising on compliance. With NVIDIA and Amazon Web Services (AWS), BRIA AI increased training and production efficiency.

Given that BRIA’s training data comprised hundreds of millions of images, optimizing data management and access was critical. To stream massive datasets to GPUs without slowing down the process, BRIA required a highly optimized storage and distribution system. 

Efficient Data Management in the Cloud

BRIA selected NVIDIA and AWS to help maximize GPU efficiency and minimize the costs of its training process. This included building a scalable machine learning (ML) pipeline using NVIDIA NIM, Amazon Simple Storage Service (Amazon S3), AWS Lambda, Amazon Simple Queue Service (Amazon SQS), and Amazon SageMaker AI. This approach eliminates the need to download full datasets onto local storage, reducing latency and improving performance for large AI workloads. BRIA also distributed training across high-performance servers powered by NVIDIA H100 GPUs. With high-performance training solutions, BRIA empowers customers to go further with generative AI while maintaining compliance.

“NVIDIA has been instrumental in optimizing BRIA’s generative AI models. Their support has enabled us to deliver commercially viable visual AI solutions to our customers.” - Misha Feinstein CTO, BRIA AI

Outcomes

Try Bria 2.3 on NVIDIA


To see all our source-available models, ControlNets, and LoRAs, check out our model page.