Watch the video:
Timestamps:
0:00 - Intro/Explanation
0:21 - Safetensor vs ckpt models
1:39 - Install Model Converter extension
2:08 - Convert ckpt to safetensor models
2:39 - Training a model in the future? Choose these.
2:50 - fp16 vs fp32
3:32 - Why not bf16
3:40 - Pruning: disabled vs. no-ema vs. ema-only
4:17 - Image quality: disabled vs no-ema vs ema-only
5:20 - What you should choose when converting models
Stable Diffusion models taking tons of space? You can compress them by pruning them. Odds are you can shave off tons of GB, and still keep crystal clear quality for your models. Assuming you’re not going to train or merge those models in the future you can ‘compress’ them even more! This guide shows you how to convert to safetensors, fp32, fp16, no-ema and ema-only models, as well as explains what you should pick and why! It’s everything you need to know in one video.