How Mixbook leveraged genAI to offer personalized experiences
summary
Mixbook combines Amazon SageMaker, Amazon Bedrock and Amazon Rekognition with Mission Cloud to auto-curate photos, generate backgrounds, stickers and captions and assemble print-ready layouts. The team builds multi-model pipelines that chain vision outputs into LLM prompts, enforces image-resolution checks and upscaling, and balances automatic theme selection with template/content mismatches. Experiments find a 7B model outperforming larger models on captions and the project spends 60–70% engineering time on bias and toxicity and on LLMOps, infra and token-cost tradeoffs.
AI-written summary · about
chapters (16)
- 0:00Introduction to Mixbook
- 1:06Mixbook's mission and products
- 2:15Mission Cloud overview
- 3:19GenAI for photo book creation
- 5:48Building GenAI systems
- 7:53Testing captioning models
- 9:20Model size vs. performance
- 10:46Photo clustering challenges
- 12:38Prompt engineering for captions
- 15:33Key findings on prompts
- 17:17Bias and toxicity guardrails
- 19:05Production challenges
- 20:47SageMaker JumpStart benefits
- 21:29Multi-model systems
- 22:32Future GenAI features
- 23:19Call to action and offers
related talks
30mBring AutoML to all users across your business
41mKeynote
39mBuilding and running an AI powered product
47mBuilding ML practices to address the top four use cases
42mHow to Build Smart Applications Without a Machine Learning Background
25mPersonalizing Your Images with AI-Powered Features
34mBuilding Owly: An AI Comic Video Generator for My Son
37mEnhancing Media Experiences in the Age of AI: A Practical Guide