USE DANBOORU STYLE PROMPTING

--------------------------------------------UPDATE-------------------------------------------------

V2 detailed is out, this new version is not necessarily more detailed than V1, the main improvements are actually in the generalization and creativity, so you should be able to get more interesting results with this version.

------------------------------------------------------------------------------------------------------

Quick comparison of V2 vs V1 detailed:

Quick comparison of the 3 models (v1):

MOST IMPORTANT TRIGGER WORDS FOR EACH MODEL:

Mixed: 3d, realistic

Detailed: 3d, realistic, detailed

Smooth: 3d, realistic, smooth

For V2 of the Detailed version in some cases adding the "photorealistic" trigger can help as well (especially with faces on anime characters), it is optional though, generally speaking the plain "3d, realistic, detailed" triggers give the most consistent results.

A little more details about each version:

• "mixed" version is a merge of 2 models I trained, one on UE5 content and another on less realistic 3D stuff. We still get some realism but I wanted to keep this mostly on the CGI/3D side, the main goal is not realism.

• "detailed" is a finetune of the mixed version, it was trained on a cropped in version of the original dataset, this results in more fine detail and more realistic shading. V2 was trained on its own with a new dataset.

• "smooth" is a merge of 2 failed training attempts (one under trained the other over trained), the result is smooth skin and more stylized look with exaggerated features