A LoRa trained on 4225 images from the mobile game Fire Emblem Heroes.
Works well at 1.0 weight, mixes with other LoRa as well, see sample images.
I would like to give a huge thanks to Lazy on the discord for all their help with DAdaptation and LoRa training in general. Check out their guide: https://civarchive.com/models/77908/lora-lazy-dadaptation-guide
The LoRa generally produces art with a style like "full body, standing, black background"
My prompt for the sample images was straight forward:
Prompt:
masterpiece, best quality, <lora:FEH_Concept_v2:1>, black background, simple background, 1girl/1boy, standing, full body, etc...
Negative:
(worst quality, lowres), blurry, bokeh, depth of field, error, text, speech bubble, artist name, signature, border, sketch
I primarily used Hassaku model (NSFW) https://civarchive.com/models/2583?modelVersionId=62528, but any NAI-based model should work.
The magical particles can be somewhat controlled, see sample images.
Weapons are very hit or miss.
You can also generate 2 people, 2girls/2boys.
v1.0 had problems so don't ask about it (you can see in the sample images)
Description
FAQ
Comments (7)
Absolutely amazing stuff and a ton of fun to just play around with.
so true, oomfie!
I can even hear the FEH recruitment theme when I make an image with this. Great LORA.
haha glad you like it!
Did you use 4000+ images to train this Lora? 😮 If I'm understanding this correctly. Because I made some Lora models too, using 20 images or so.
This particular LoRA yes, i've made LoRA with as little as 4 and as many as well.. 4225.
@richyrich515 wow omg! That's a lot of work! Good job 👏🏻 And thank you so much for making it. I've made several images with it, it's really awesome. Is there any chance that you would make it for Pony XL too? ^^
Details
Files
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.