How to Make Stable Diffusion Generate Not Just Anime and Cats, but You
summary
Overview of methods for personalizing Stable Diffusion-style text-to-image models so they can generate specific people rather than just common subjects like anime or cats. Covers recent academic and production approaches from 2022 onward for fine-tuning or conditioning generative models (Stable Diffusion) to capture a particular person's appearance.
AI-written summary · about
chapters (19)
- 0:00Introduction
- 1:07Realistic image generation problem
- 2:58Diffusion models basics
- 5:05Stable Diffusion advantages
- 7:12Personalizing model for a person
- 9:25Training data requirements
- 11:17Choosing training models
- 12:39Textual Inversion method
- 14:45DreamBooth method
- 16:11LoRA optimization method
- 18:42Training methods comparison
- 20:07Model training optimization
- 21:32Latent representation caching
- 22:58Chain Rule memory optimization
- 25:09Attention mechanism in models
- 31:53SoftMax optimization for attention
- 32:15Optimized model results
- 32:38Summary and applications
- 34:04Q&A
related talks
10mDiffusion models
44mGenerative Adversarial Networks (GANs) and Stable Diffusion
41mBuilding Generative Image & Video models at Scale
29mIntegrating Stable Diffusion Generative AI in Artist Workflows
01h 21mGenerative AI with Diffusion Models
52mThe Yoga of Image Generation with Stable DIffusion and ComfyUI
01h 27mImage Generation with AI
47mFuzzy Humanity Put into Production