Fuzzy Humanity Put into Production
summary
Shows how the presenters trained Stable Diffusion on photos of 60 employees and deployed a real-time system that lets coworkers send prompts in Slack, receive four generated images, pick favorites and post them back to the channel. Describes technical lessons on dataset curation and training parameters for combining many people into a single model, and discusses societal issues such as biases, copyright and social impact of generative algorithms.
AI-written summary · about
chapters (11)
- 0:00Introduction to project
- 3:36Birth of Grafita
- 6:30Diffusion models explained
- 9:43Training with employee photos
- 13:18Personalization with DreamBooth
- 18:16Evaluating generative models
- 25:45Lessons learned
- 33:18Privacy and data authorship
- 35:03Fuzzy humanity in AI
- 37:16AI future and challenges
- 38:07Q&A
related talks
01h 03mIt's All About AI
58mWorkshop: Empowering Collaboration: AI Developer Experience - Your Bridge from Mod...
56mPicasso's terminal; data science and AI in the visual arts
49mAstronaut Horse: Stable Diffusion Artist Collaborations
41mAdding Generative AI to Real Time Streaming Pipelines
46mHow to Make Stable Diffusion Generate Not Just Anime and Cats, but You
01h 27mImage Generation with AI
47mThe impact of Generative AI in different industries