Best setting for training SDXL Loras

[ShinobiiSpartan]

Hey everyone, hope all is well, I have a question, what are the best settings for training loras on the SDXL Google collab to where I'd get a close 1:1 of my character? If anyone can help, I'd really appreciate it.

[hollowstrawberry]

The default settings are usually fine, though you have to adjust the repeats depending on how many images you have. The tagging of the images is more important I'd say. If you're using Pony, it needs a lot more tags compared to sd1.5 but in turn achieves higher fidelity.

[ShinobiiSpartan]

The default settings are usually fine, though you have to adjust the repeats depending on how many images you have. The tagging of the images is more important I'd say. If you're using Pony, it needs a lot more tags compared to sd1.5 but in turn achieves higher fidelity.

Sounds good, thanks a lot. If possible, would you be able to quickly give me an example of how many repeats you'd use if let's say I have 165 images?

[hollowstrawberry]

I would personally base it around 1 or 2 repeats and 10 epochs. 165 images is enough that you might not need more than 1 repeat. If it was 1.5 you could try as much as 3 repeats.

[ShinobiiSpartan]

Ok, thanks a lot as always and have a great rest of your day or night🙏