Inzaniak

Let's talk about SDXL!

Added 2023-09-23 22:26:08 +0000 UTC

Hi everyone!

Today, I've spent most of my day playing with some Stable Diffusion XL Anime models to see how they performed.

Here are a few things I've noticed:

Generating pictures using SDXL models has improved, but it still takes way more than using 1.5 even when generating the same resolution.
Smaller details are usually more defined when using SDXL
I find it harder to pinpoint a good resolution for generating the base picture. I know that SDXL works with 1024x1024 pictures, but I can't find a sweet spot for vertical pictures.
I've tried training a LoRA, and it seems to work flawlessly. Not only that, but I've trained one over some pictures made with Mistoon_Anime_v2, and I'm going to share it with my subscribers in the next few days (it's not that great, but it's a nice experiment)
I made a few pictures with the LoRA and compared to the pictures made with 1.5, and I must say that right now I don't think that working with SDXL is worth it for anime pictures. You can find an example attached in this post.

I'd like to know from you if you've used SDXL and what you think about it. Do you know a good Anime model? Do you use a specific workflow? Answer to the poll and let me know in the comments below!