Resource :: Model training
Model training is the process of teaching a Stable Diffusion model to understand visual concepts and the relationships between them.
The term includes the base model training, fine-tuning (IE generating a checkpoint, see the term), using Dreambooth (not used casually anymore, replaced by LoRA) or training a LoRA.
We'll focus on LoRA training.
LoRA training
Everything you need to know is available on the GitHub page.
Stable Diffusion Art provides an excellent guide, for SD1.5 and SDXL.
Essential to Advanced Guide to training a LoRA on CivitAi, more advanced and suitable for experimented users.
How does Numeratic train lora's? on CivitAi, made by the master of anime LoRAs. It is worth the read but I disagree with the tools used and
prefer Kohya_SS. The most useful part is about Grabber, you will use it too. A lot...
Explaining LoRA Learning Settings Using Kohya_ss for Stable Diffusion Understanding by everyone on GitHub, extensive guide made by the author himself. A must-read.
You can also ask Gemini.
Where to train a LoRA?
Very simple: the easier the most expansive and the less customizable.
Paid services (beginners)
CivitAi If you plan to use a paid service, supporting CivitAi is a good idea.
PixAI Not tested.
Paid services (advanced)
Google Colaboratory provides the hardware and some help. Useful if you don't own a modern graphics card.
Free (expert)
Can you use it with a low-end graphics card? No.
Is it fast? Oh no...
Are the settings intuitive? Or well explained? Is there an automatic install? An automatic updater? Oh boy...
It is easy to use? ABSOLUTELY NOT!
Should I use it anyway? Do you see a world where knowing how to train an AI is useless?
Kohya's GUI for Kohya's Stable Diffusion trainers.
Can train Textual Inversion, LoRAs or DreamBooth, fine-tune, convert/merge models, and comes with many utilities like auto tagging (BLIP, CLIP, WD14...).
Kohya SS is so ahead of all the other tools that there is no point at all looking for another one.