Custom Models for LoRA Training in the Colab

Suppose you're training a LoRA with the Colab by Holostrawberry, and you're using a custom LoRA. If you were to, instead of the three listed in the Colab, use the custom URL and right click on the download from one of Civit's files, (i.e. doing this link: https://civitai.com/api/download/models/128713), would that in turn use that model for the training or would it just not do it?

I only used this option with huggingface links and that always worked flawlessly. You should be able to see that it tries to fetch the model once you start the training. It's gonna take a minute to download the model for the remote training.

Gave it a try:

it's downloading the model so it should work.

Just be aware that often training on one of the base models is more accurate and flexible. Only add a model when training for yourself and you only want to use that checkpoint.

I made the mistake of immediately training on AOM3. The models didn't work with anything else and were strictly worse than the results I got when retraining on animefull.