A significant improvement over the last version, most likely the last release.
More consistent faces at a distance
Less human skin issues
Less artifacts and mutations, more successful generations
Does not force realistic style as much
Usable across wider range of checkpoints, just experiment with lora weights
While it works much better with anime models now, I still recommend using the dedicated anime releases for Anything V4 and AOM, as the loras there are more responsive, but this is a good general model.
Style and Models: Works well with the most checkpoints, recommended weights for the ones I tried. All examples are straight txt2img
DeliberateV2 (0.6-0.9)
Realistic Vision (0.6-0.9)
RPG (0.7 - 1)
Dreamshaper (0.5-0.8)
Chilloutmix (0.8-1)
DalcefoeV4 (0.7-0.9)
Dalcefoe oil paint V3 (0.7-0.9)
DalcefoePaintingV4 (0.7-0.9)
AOM (0.8-1.0)
Anything V4 (1.0-1.2)
Anything V3 (1.0-1.2)
ExpmixLinev20 (0.9 - 1.0)
Settings: Works best at low to mid CFG (3-6), I personally prefer to pick lower end of the scale.
