Somewhere in the out rim of the galaxy lies a forgotten colony. Its inhabitants are former humans, who continously enhanced their bodies with synthetic part to survive in this wasteland. Their behavior has changed nature as well.
Don't forget to leave a ❤️ to get a notification for future updates. ⭐ reviews are appreciated, too. Let me see what you created with this tool. Follow me, if you don't want to miss any new stuff I might create. Enjoy the possibilites of this model.
Use the triggerword and synth for the real effect. I think a lot of stuff got absorbed into synth. ;)
Be aware of witches, husks and cultists. They can be pretty scary.
Use prompts, to enhance the overall feel or push the LoRA into different directions.
Scifi:
use scifi, tech, futuristic, futurism to get it even more techy
Fantasy/magical:
use magical, whimsical, fairytale, fantasy
Darker:
eerie, stygical, gloomy, decay, haunting
For more inspirations check out my little GUIDE.
The samples were created using my Wildcards.
weights around 0.8 work fine
Description
FAQ
Comments (11)
In my opinion, your best so far. ;)
thx ;), It's ~380 input images oO
@DonMischo I don't know if more the better. Probably around 60 should be a good spot but that's just me. Good work mate.
@LDWorksDavid I can't decide sometimes ;)
Nice. I've a question: when training LoRA for SDXL, do you set "Text Encoder learning rate" to zero?
Hmm, I don't think so. I'll check later. Or you could check in A1111, with the small ! When mouseover a Lora, you'll get all training settings
@DonMischo Oh, ok, I see there "1.0" for both learning rates, u-net and text encoder. These values are strange.
@konyconi I'm not near my computer right now. But I think it's related to the optimize arguments. I think I followed Faeai und Alles Guide at education.civitai.com
@DonMischo I'll check them, thank you.
@konyconi If you are using any optimizers besides Adam then setting the rates to 1 is to ensure the optimizer can actually just do it's job! :)
@olivetty Aha, thank you.