santa hat
deerdeer nosedeer glow
Sign In

LoRAs For NoobAI XL 1.0 Epsilon announce + optimal settings

LoRAs For NoobAI XL 1.0 Epsilon announce + optimal settings

Hello, my Followers!

Intro

I want to announce to you the release of new versions of my LoRAs for another SDXL model - NoobAIXL (Epsilon 1.0).

I will upload them gradually as they are ready. The first 3 LoRAs will be uploaded today!

I don't recommend to use V-pred now (only for enthusiasts) because it is not finished and LoRAs for Epsilon and Illistruious are not compitable with V-pred version! If V-pred will be good I'll make LoRAs for it too.

According to my tests it is very good in anime and cartoon style. Unfortunately I didn't find good settings for 3D realistics like you saw on my images with Overwatch and WoW girls. But I think it's a matter of time and soon we will see 3D mixes based on NoobAI like Blendermix.

Contrary to the misleading name NoobAI is not a simple model for noobs. It doesn't understand natural language unlike PonyXL and you have to use only danbooru and e621 (furry) tags.

Also you need to use right settings or your image will be over-contrasted!

However, I have tested the model and found good settings that will make your pictures on the level of PonyXL or even better!

I also continue to make LoRAs for PonyXL as well! It is not replaced!

I train all my LoRAs on NoobAIXL v0.75, but I use NoobAIXL 1.0 because LoRAs trained on NoobAIXL 1.0 are very contrasty. According to my tests NoobAIXL 0.75 loras are better than 1.0 and IllustriousXL loras.

I don't see any sense to train LoRAs on IllustriousXL since Illustrious is worse than PonyXL and there is only one good finetune of IllustriousXL - NoobAI-XL because it is finetuned with 12 million arts of danbooru and e621 (HUGE! PonyXL has only 2.5 millions arts but also 3D).

BUT! If you need IllustriousXL versions I can also upload them.

Optimal settings for NoobAI-XL

I use classic NoobAIXL or NoobaiCyberFix (a mix based on NoobAIXL in which the base model of SDXL 1.0 (which is finetuned by Illustrious and NoobAI) is replaced by Cyberrealism XL, which has no style of its own but has better anatomy according to description). I hope we will see other good mixes soon like Autismmix for pony.

My cfg 4-5. Higher cfg's cause over-contrast.

euler / euler a with 25-30 sampling steps.

Quality prompt (very important):

Put "masterpiece, best quality, good quality" in positive and put "low quality, worst quality" in negative.

If you have over-constrasted picture and cfg 4-5 didn't help you can try to put in negative "colorful".

You don't need to use artist tags if you use my character loras or style loras but if you want to use it check danbooru for artist tags!

Also for everything (poses, body, faces, actions, backgrounds, etc) use danbooru wiki (for anime and everything) and e621 (for furry tags).

18

Comments