About this version

Every training was done with NL and tags, but only for style purposes.

Tested with my settings:
22 to 28 steps, CFG 3.5 to 4, er_sde, no artist tags and minimal quality tags.

Better:

• Better background compositions, less random, more structured
• Fewer "melted-looking" elements

Slightly better (in around 1–2 out of 5 images):

• Better hands
• Fewer random light particles

Better or worse depending on taste:

• Faces and bodies are more in shadows
• It tends to produce fewer simple "dot noses"

Note: No additional knowledge was trained into the model, as Anima have severe knowledge loss across concepts when trained for extended periods without datacenter-level hardware, which I do not have.