Trained on the Civitai On-Site LoRA Generator!
The WHAT!? You heard! We are testing an amazing on-site LoRA Generator which will be launching soon to our awesome Civiai Supporters!
Tell me more!
I will! So we're trying to make it as easy as possible for people who've never trained before, but also be fully-featured enough for power-users to want to use. I have an RTX 4090, but the results I get with the on-site trainer are so good... why should I use my own electricity to train? 👀
It trains 1.5 & SDXL?
It sure does! Depending on whether you choose a Style, Concept, or Character, it'll give you pre-sets based on the number of training images uploaded - and it even does captionless training - this model was 17 images, ZERO CAPTIONS, default settings 🤯
Find out more about our LoRA trainer here - https://education.civitai.com/using-civitai-the-on-site-lora-trainer/
So, Empire?
Empire Style was one of the first pieces of content I uploaded to Civitai back in December of 2022. It quickly became popular, adding epic city streets, cobbles, beautiful backlit scenery, and an overall kind of "golden hour" glow and depth. I've expanded the data set a little, and here we are for SDXL - doing similar, but better!
Description
FAQ
Comments (11)
I was just thinking of the 1.5 version and wanted to follow up my western series on XL. This should totally help.
I need the Egyptian back ^^
Gonna do it!
@theally and I totally forgot about the princess ^^ Dunno how much time I wasted figuring out why they did not work on the model... yeah... using 1.5 on SDXL is kinda pointless ^^
@sheenariel Well, Textual Inversions work at half-strength on SDXL due to the partially shared architecture, so they're still somewhat useable.
@theally so does that mean, I could try to use the old ones and just amp up their strength? Like dialing the standard 1 up to 2.
@sheenariel Absolutely! Should work :)
@theally this doesn't seem right. I'm using Hamunaptra here, and at even 0.05 strength it just completely ruins the results. Here's an example with Pikachu; all settings including seed are exactly the same other than the introduction of the Hamunaptra embedding. I can upload the direct results from Horde if you'd like, it should include a reusable ComfyUI workflow embedded in it. Using longer prompts seems to have similar results.
No TI: https://tinybots.net/artbot?i=SrQCTEURAOu
TI at 0.05: https://tinybots.net/artbot?i=VY4Yc35WUog
I've uploaded the resulting image from horde at https://civitai.com/images/6393963
I should note that ArtBot instructs Horde to apply the embedding to the prompt automatically. I can place it manually, but not using ArtBot.
could you share the training data?
Wow! Nice.
I wish there was a Flux version of this.
such a great LORA, love this one!