A Video Retrieval Augmented Generation (VRAG) pipeline built on AWS that generates custom videos by combining user-provided text prompts with images retrieved from an indexed dataset. The solution uses Amazon Bedrock, Amazon Nova Reel for video generation, OpenSearch Serverless for vector search, and Amazon S3 for storage. Seven sequential Jupyter notebooks walk through image processing, vector ingestion with Titan Embeddings, text-only video generation, text+image video generation, multi-modal VRAG, in-painting, and combining in-painted images with video generation. Use cases include educational videos, marketing content, and personalized media. Deployment is via CloudFormation in a SageMaker notebook environment.

13m read timeFrom aws.amazon.com
Post cover image
Table of contents
Solution overviewExample inputPrerequisitesDeploy the solutionRun notebooksBest practicesClean upConclusion

Sort: