Consistent Storyboards with Amazon Nova: A Guide

Amazon Nova Canvas and Reel, AI foundation models within Amazon Bedrock, are revolutionizing storyboard creation. This two-part series focuses on achieving visual consistency, a challenge with AI-generated visuals. Part 1 addresses prompt engineering and character development. Key techniques include detailed text prompts specifying character attributes and clothing, separating style descriptions (e.g., “graphic novel style”) from style details (e.g., “bold linework”), and using the seed parameter for character variations while maintaining prompt consistency. The cfgScale parameter controls prompt adherence, balancing creative freedom with consistency. An end-to-end storyboard pipeline is presented, leveraging Amazon Nova Lite for prompt optimization and Amazon Nova Canvas for image generation. This pipeline generates multiple image variations while maintaining consistency across scenes. For animation, Amazon Nova Reel is integrated with Amazon Nova Lite to create short animated video clips from image prompts. While these techniques significantly improve consistency, Part 2 will explore fine-tuning for near-perfect results. The target audience includes filmmakers, animators, advertisers, and UX designers involved in pre-production workflows. No specific technical specifications beyond the use of Amazon Nova Canvas and Reel, and the parameters seed and cfgScale, are mentioned.

(Source: https://aws.amazon.com/blogs/machine-learning/build-character-consistent-storyboards-using-amazon-nova-in-amazon-bedrock-part-1/)