Type | |
Stats | 3,987 96,688 |
Reviews | (239) |
Published | Mar 4, 2023 |
Base Model | |
Training | Steps: 2,300 |
Trigger Words | protogen, protogen visor, protogen face |
Hash | AutoV2 9B0FB125DF |
A LoRA trained on many images of Protogens in order to get them as accurately as can be, what else can be said?
IMPORTANT: The AI will always struggle with the visors first time, they are very easy to inpaint, here is how:
Send image to inpaint. Enable LoRA in the img2img tab as well. Set denoise strength anywhere from 0.65 to 0.85, adjust accordingly. Make sure the positive prompts only have "protogen, protogen face, protogen visor" in them and nothing else. Mask only the visor portion. Make sure setting is on original, and masked image only.
Activation keywords are: protogen, protogen visor, protogen face
Features three different versions:protogenv2-hard
- Trained on 3200 steps, more "canonically" accurate to proto designs and training dataprotogenv2-medium
- Trained on 2300 steps, might be more lenient towards nsfw and more creative designs, includes a lesser trained epoch as wellprotogenv2-soft
- Trained on 1900 steps, may be even more creative and lenient, however it may not always be the most correct