Building Owly: An AI Comic Video Generator for My Son
summary
An end-to-end system generates personalized comic videos from a title and a few toy photos using Amazon Bedrock LLM to write multi-paragraph scripts and a fine-tuned Stable Diffusion 2.1 on SageMaker JumpStart to keep a consistent toy character; image fine-tuning uses only handfuls of photos, training and deploying a new endpoint, then a pipeline (S3 + EventBridge) runs batch renders, Amazon Polly narration and music selection, and stitches with moviepy — it works reliably for nightly bedtime stories.
AI-written summary · about
chapters (19)
- 0:00Intro
- 2:04Project faAi
- 2:56Project Ellee
- 4:20Demo: Ellee in action
- 6:12Owly – a personalized comic video generator
- 7:44What does Owly do?
- 8:09Demo: Owly in action
- 10:49How does Owly work?
- 12:58LLM
- 13:18Demo
- 16:52Amazon Bedrock
- 17:10Building the comic image generator
- 17:45How the Stable Diffusion model was built
- 19:52Challenges & enhancements
- 24:28Problems & solutions
- 26:44unplanned cool feature
- 30:10Architecture
- 32:15Something cool
- 33:48Outro
related talks
12mAI Content Generation
13mSpotting and solving everyday problems with machine learning
39mBuilding and running an AI powered product
50mBuilding And Running An AI-Powered Product
45mGenerative AI for Image and Video Production
09mRethinking AI: Building What Matters, Not Just What's Possible
48mHow I became my son's hero by building an AI Minecraft Mod with LangChain4J
43mBuilding Qrio, a Bot That Plays Videos for My Toddler