V2.7 IL/NOOB Version
skipped 2.0-2.6 as these where failed versions
Suggested weight anywhere from 0.8 to 1
Scheduling type depends on base model you use but in most cases Euler a works fine.
this LoRA was 100% made with my own self-made art, I rendered everything in Blender cycles/eevee
This version was trained locally on my pc using the same dataset as last version but it has way better results and the file size is much more agreeable
Description
V 2.7 release
(v 2.0-2.6 where failed iterations)
trained locally using comfyui on a rtx 4070
using same data set as last model (115 images)
Repeats 4, batch size 2, training set at 0.0005 @4000 steps
Best used with noob/IL based models (I use yiffymix)
Use 1-0.8 for 3D and 0.7-0.6 for cel shaded look. anthing after 0.5 is barely noticeable.
I am the artist who made the training data (I trained my own style off my own work made in Blender)