Watch the video:
Timestamps:
0:00 - Intro/Explanation & What is a LoRA
1:15 - Installing Kohya_ss
1:54 - What we're selecting
2:20 - Selecting a model to extract a LoRA
3:13 - VRAM/RAM
3:27 - Extract LoRA from Model (Convert Model to LoRA)
4:13 - Model & LoRA conversion size (From 4GB to 600MB)
5:00 - Importing new LoRA & model to AUTOMATIC1111
5:50 - Generating images with Model & LoRA + Another model
7:38 - Direct comparison from Model to SD Base + LoRA
8:30 - Why create LoRAs from models?
8:42 - Using one model's style in another (using extracted LoRA)
9:40 - Why not merge models? This is better.Like the aesthetic of a model, or does it get certain things REALLY right? This guide shows you how to convert a model into a LoRA, allowing you to take a 4GB+ file and shrink it down to a few hundred MB, while keeping all the great aspects about it! This is incredible! I also haven’t seen guides on this, so here’s to a first quality guide!
One-click Kohya_ss install: iex (irm kohya.tc.ht)
Guide to using the above command; Kohya_ss install: Go to post or View on YouTube
AUTOMATIC1111 install: Go to post or View on YouTube
