Introduction
Hey everyone! :)
This is a LoRA of the Character "Dokkaebi" from the game "Tom Clancy's Rainbow Six Siege"
The model was trained on Stable Diffusion 1.5 with 30 images picked by hand from in-game and different trailers.
Recommendation
The Keyword "RSDokkaebi" is not necessarily needed but helps to get better results. Words like Uniform, Beanie, Round Glasses should also help.
a weight between 1.0 - 1.2 is recommended (take a look in example images)
Models used in examples
DosSkin02Koto04: (Just an experimental merged model = DosMix & PoV Skin Texture - Dreamlike r34 with a multiplier of 0.2 and after that merged with Kotosmix with a multiplier of 0.4)
Examples
All examples were created with Hires. fix from 384x512 Upscaled by 2 or 2,5 (or 512x512 to 1280x1280)
Hires steps: 30
Denoising strength: 0,6
Hires upscaler: Latent (nearest-exact)
Model used for example is mentioned at the end of negative-prompt
Description
FAQ
Comments (8)
If you would consider Fury from Ghost Recon :)
What would really be valuable is to get the style from all the operators
My plan is to create a seperate lora for every operator. If someone can confirm things does not get mixed up and stay clean, when all operators are packed in one lora, I would try to do so
This looks great and i will try using this at some point. Could you tell me how you trained this lora?
What I did is taking some screenshots from the in-game operator from different angles and distance. In addition to that I added some cosplay and nsfw images of the operator from different sides. Cropped them to 768x768 and edited the BLIP-Captions to my liking, so they fit more to the operator.
For the training itself Im using the Kohya_ss GUI. Left basically all settings on default, except for max resolution (768x768). Trained on batch size 1 with 300 steps I think.
I hope this answers your question :)
@Kalibi it does thankyou!
you forgot watermark
Can you add Lore of Aruni from Rainbow Six Siege?