MENU

Fun & Interesting

Make YOUR OWN Images With Stable Diffusion - Finetuning Walkthrough

Adam Lucek 19,472 10 months ago
Video Not Working? Fix It Now

What’s better than an image generation model? One that can generate YOUR specific images! GUIDE SHOWN IN VIDEO: https://docs.google.com/document/d/1KRHOdVA75olgOQNJOSlI29OYkfh-zTTXwaEssPlZ8rA/edit?usp=sharing In this video we break down fine tuning Stable Diffusion for two specific tasks: Style Transfer - Taking a dataset of public old book illustrations and training so that all other Subject Permanence - Using Google’s DreamBooth research to generate a picture of my apartment chair in different scenes and colors We cover all of the background of text to image diffusion model training from the top, breaking down the complex topics to better understand how and where we come in to effectively train on our own data. Resources: Intuitive Breakdown of Diffusion Models: https://youtu.be/4iR_odrRT8c Old Book Illustrations Model: https://huggingface.co/AdamLucek/sdxl-base-1.0-oldbookillustrations-lora Old Book Illustrations Dataset: https://huggingface.co/datasets/AdamLucek/oldbookillustrations-small Old Book Illustrations Colab Notebook: https://colab.research.google.com/drive/1SMLhqfW2xjOK8Ep3R2FGT-xUyX50DgZh?usp=sharing Green Chair Dreambooth Model: https://huggingface.co/AdamLucek/sdxl-base-1.0-greenchair-dreambooth-lora Green Chair Dataset: https://huggingface.co/datasets/AdamLucek/green-chair Green Chair Colab Notebook: https://colab.research.google.com/drive/1v503hMrThIy87xozZBMPSDBc53Bukk_1?usp=sharing HuggingFace Diffusers Package: https://huggingface.co/docs/diffusers/en/index Chapters 00:00 - Intro 00:50 - Understanding Stable Diffusion 05:56 - Methodology: Text-to-Image Model Training Overview 12:02 - Methodology: Low Rank Adaptation 19:00 - Methodology: DreamBooth Text-to-Image Training 25:57 - Methodology: Additional Methods & Ideas 29:24 - Prep: HuggingFace Diffusers Package 31:48 - Prep: Diffusers Fine Tuning Scripts 32:58 - Prep: List of Script Arguments 35:14 - Prep: Specific Text-to-Image Script Arguments Breakdown 43:03 - Prep: Specific Dreambooth Script Arguments Breakdown 43:47 - Training: Finding a GPU 45:55 - Training: Setting up GPU Environment 49:25 - Training: Running the Script! 52:07 - Outcome: Load & Run Style Tuned Stable Diffusion 54:23 - Outcome: Load & Run DreamBooth Tuned Stable Diffusion 57:02 - Outro #ai #stablediffusion #programming

Comment