Nevolution

FaeTastic family of checkpoints is great, but in some cases it lacks diversity of results. This checkpoint (Nevolution) is an attempt to mix in some checkpoints with other styles and effects. A few checkpoints (including in particular DucHaiten-AIart) were taken as initial ancestors and were mated with secondary checkpoints from other "breeds" (Pilgrim Midjourney for visual effects, DucHaitenNiji for some artistic effects, etc). Then, there descendants were mated further and further. Selection took place to reject offsprings producing artefacts. And some other checkpoints (both base and merges, including, e.g. Animagine) were merged with low weights for tweaking the result, which proves itself to be good enough.

The checkpoint is general-purpose, but it was asserted mostly on sci-fi / cyberpunk / fantasy content.

Version 1.0 is the initial result of selective breeding of multiple checkpoint. This version provides more stable, but slightly less diverse results.

Version 2.0-base is the version, which was obtain form breeding Version 1.0 with a number of more recent models (e.g., realismFromHadesXL, PixelAlchemy, NovaAnime, etc.) These models were taken with 0.05-0.1 weight, so it's still the same Nevolution model, but somewhat more versatile, and in some cases more detailed.

Version 2.0-anim is a sister of 2.0-base. These checkpoints share all the parents, but with different weights. Their "genomes" are similar, but different. 2.0-anim version got higher contribution from illustration-oriented parents. To amplify its capabilities, such words as "anime", "illustration", "manga", "painting", etc. are better to use. The checkpoint is still quite versatile and not specifically anime-oriented, which may let it produce unique results.

Version IL-v1 is an independent merge of Illustrious checkpoints. However, I tried to make the resulting checkpoint to work similar (up to the difference of the base models) to other Nevolution checkpoints. So, it's a kind of convergent evolution.

Recommend Settings

Sampler: I typically use DPM* or Euler a, but others also work

Steps: around 30 for v1.0; 30-45 for v2.0, 30 is enough for IL-v1

CFG Scale: 5~8