High-Res-Fix and Dynamic Thresholding
This model will no longer be updated as I feel like its starting to regress, I will be training LoRA's from now on. If you have any ideas for LoRA's let me know. I only have an RTX 2060 6gb so they might not be the best, the max resolution that I can train is around 600x600.
This model's goal is to generate photo like images.
Inpainting version on HuggingFace
All images where generated using High-Res-Fix with 4x-UltraSharp upscaler.
vae-ft-mse-840000-ema-pruned
Description
This will most likely be the last version, as I am satisfied with the state of this model.
This version should give better armor, and improve the realism that was lost in the last version.
If the effect of the skin LoRA is too much, try playing with the weight of the trigger words.
Fantasy skin colors (red, blue,green), seem to be hit or miss, from my testing you will see more success if you add words for other things that also have that skin color (ogre, avatar)