Introduction (Check Out Versions Before Downloading!)
This is my first LoRA. I barely know what I am doing. Pls don't be rude. It's based on the best girl Ryuu Lion from danmachi. I used various fanarts and anime screenshots in training. It's flexible and you probably won't get the classic look of the character due to training images. I recommend using the LoRA at 0.5-0.6. Also Clip Skip "2" is recommended but not necessary.
V2 Update:
This was trained differently with the same images. You can call the character only with "ryu, elf". You can also use detailed prompts like "ryu, 1girl, elf, blonde, short hair, wavy hair..." to get photorealistic results in models like chilloutmix. Though somehow the first model I randomly trained is better at photorealism in these models.
Description
Smaller filesize.
Different training method.
Less flexible than the first version but more consistent.
FAQ
Comments (2)
The demo images are stunning. how long did it take to train this Lora? how many images did you use?
Hey! Thank you. I just pulled around 30 images from various sources, then tagged and cropped them myself. I think the secret sauce is data variety, optimal amount of epochs/steps and noise offset. I don't remember the exact settings unfortunately but there wasn't anything interesting other than the stuff I mentioned.
Details
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.