The Yoga of Image Generation with Stable DIffusion and ComfyUI
summary
Project uses Stable Diffusion + ComfyUI on a local PC to produce consistent yoga images for YogĀrkana, combining ControlNets for pose (depth/edges), embeddings and LoRA for style, and FaceID and face detailer for stable identities; training a custom LoRA with ~15–28 photos and ~20 epochs proved slow and low quality, so runtime FaceID is used instead. Workflow yields correct poses and repeatable faces; a multi-frame video prototype with One model runs but takes many hours. embeddings/LoRA chaining enables stylistic variations.
AI-written summary · about
chapters (17)
- 0:00Introduction to project
- 1:33Why Stable Diffusion
- 7:32First workflow demo
- 19:24Limitations of text-to-image
- 20:06Image-to-image workflow
- 22:35ControlNet for pose control
- 27:10Brand identity challenge
- 28:13Using LoRAs for faces
- 33:35Training personal LoRAs
- 37:27Face ID and face detailer
- 42:20Final workflow demo
- 43:43Comparing with cloud solutions
- 46:31Science and art of generation
- 47:11Inclusivity in generation
- 47:55Website update and next steps
- 48:38Video generation test
- 50:02Conclusion and Q&A
related talks
28mBeyond DALL-E: Advanced Image Generation Workflows with ComfyUI
50mWelcome to the Latent Dystopia
29mIntegrating Stable Diffusion Generative AI in Artist Workflows
42mModeling Aesthetic Identity- Building a Digital Twin
46mHow to Make Stable Diffusion Generate Not Just Anime and Cats, but You
47mAI Image Generation: Visual Artistry with Midjourney, Adobe Firefly, and DALL-E3
47m3D Digital Avatars with Machine Learning
25mDecoding Algorithmic Artistry: Visual Generative AI