I noticed that this model is quite rigid and not really easy to work with, despite this model has good accuracy of his facial feature. So, I updated him with various wide variety image of Marcus Fenix and deleted any bad data. Here's some changes:

• Decreased image quantity from 130+ to 120+ images

  • This decrease image quantity is also followed by new addition of images that has higher quality image and varied poses and clothing.

  • Proper anime-style/flat 2D color style

  • This means NSFW image of Marcus Fenix is also properly supported

• Bit decreased steps, since image quantity reduction. But LoRA size stay same.

• Increased flexibilty of model with few or even no negative prompt for taking off his armor. This means, the model is responsive with clothings change with wide variety of styles.

v1.1:

Some of reports says that my model is generating quite a slim Marcus Fenix. When I saw that, I can see he's bit slimmer than usual. So, I drop an update for this model and some bit of experimentation. Heres the info about this update:

• Increased quantity of dataset from 60+ to 110+ images

• Increased LoRA size

• Increase image dataset's resolution from 1.9x to 4x

v1.0:

• Using 60+ images

• Epochs: 15

• Steps: 10