MOSS Image Release 10/31/2026

Suggested tags: a watercolor illustration in the style of yoshitaka amano

Suggested strength: 0.8-1.2

Same dataset as Krea 2 but trained on MOSS. Love the variation between generations in MOSS. Does a weird thing in my experience where high strength removes their arms unless you specifically describe them in the prompt? I think MOSS just does this at higher strengths. Seems way better when using er_sde bong_tangent Sampler/Scheduler.

Krea 2 Release 06/27/2026

Suggested tags: a watercolor illustration in the style of yoshitaka amano

Suggested strength: 0.8-1.2

Been having fun with Krea 2, it works quite well and knows a lot of characters. If using multiple LoRas, lower strength to .8.

Anima release 05/30/2026

Suggested tags: @amano yoshitaka, abstract, watercolor \(medium\), cover, scan, abstract background

Suggested strength: 1.0

For truest to the style I recommend Anima Base- found pretty similar results with WAI.

Loved generating the example images for this model, such a fun style. Trained mostly on @amano yoshitaka (and other tags of course) but the other Suggested tags add something that the tag alone cannot.

Illustrious release - 12/17/24

Suggested tags: amano yoshitaka, abstract, watercolor \(medium\)

Suggested strength: 1.0-1.5

Looks best on Illustrious base, gets a little more anime like with others if you're into that kind of thing. If you'd like to spice up your image there is a list of elements common to his style listed below. Also enjoyed adding tags cover, scan

Pony XL release notes:

Trained on Pony XL Base

No activation tag needed, just prompt away!

Using pony score_tags is fine, but know it will impact the style, tested with and without and liked both results.

Common scene elements you can add to help accentuate the style of the artist:

flower, weapon, jewelry, earrings, necklace, bracelet, cape, turban, hair ornament, lipstick, armor, bug, butterfly, wings, horns, dragonfly, snake, blood, helmet, headpiece, skull

Be mindful of negative prompts/embeddings, noticed my usual negatives caused predominantly orange backgrounds where there is none in the dataset.

Other information:

Compared mine to the other most popular of this style and found theirs to be a bit desaturated, which is honestly pretty accurate but I like the results of mine enough and find mine to be more flexible due to my tagging method.