Doubts regarding training my own custom Stable Diffusion model
Hi,
This might seem an irrelevant question for this thread, but please bear with me. I am looking to fine-tune the stable diffusion 1.5 model, just as you are doing for DreamShaper. However the literature available on the internet suggests using Textual Inversion, Loras and Dreambooth, and I am confused which one to begin with.
Can you just please point me in any single direction from where I could start my journey?
Big Thanks
Have you looked at how to train one instead of what to use with one?
Dreambooth is for training yes.
but you can train TI's and Loras too
What I meant was that to train a model, I was suggested to use either of Dreambooth, TI and Lora. Are these the ways through which people fine tune their models and upload on civit.ai or is there some other method also?
TI doesn't train a model, and a Lora isn't af ull model.
You can USE LORA to bind to a checkpoint
but if yo'ure looking at full checkpoint to TRAIN rathe than merge -- dreambooth
Okay, so can I assume that the checkpoint models uploaded on Civit.ai (like Dreamshaper itself) is trained with the Dreambooth process?
Not ALL of them are, like for example Dreamshape i believe is, but some people like myself then use Dreamshaper or other checkpoints via the checkpoint merge process
Honesty?
I'd say start with learning to merge first, get a feel for how you like certain styles.
THEN figure training out, because i did it backwards and now i feel bad..
Cause some of my early on stuff was just so off the wall.
thanks for the clear explanation @Duskfallcrew . Will follow your advice and learn a bit about checkpoint merging first.
The other thing is Lykon may pipe and say something different.
That's my personal opinion, and it may not be shared .
I"m not Lykon LOL, 'im just a fan of Lykons content and consider Lykon to be an AI colleague because i'm involved in a server we're both in <3
How the dream shaper model is trained , is it trained using lora, dreambooth are any other way , I am newbie in AI