One of my pet peeves with SD has always been its absolute incapability of generating something vaguely resembling a roman legionary, regardless of the model and checkpoint. So here I present my poor attempt to fix that, for all the fans of Gladiator, Sword-and-sandal films and Asterix...plus the two or three classical history nuts.

Trained on base SD 1.5 (Clip Skip 2) using 65 images combining photos and illustrations of the iconic lorica segmentata armor of the imperial period.

The base keyword is: loriseg armor

But actually the most common prompt is: wearing loriseg armor and helmet

Additional keywords: red tunic, red shirt, sandals,

shield, spear, sword

THE GOOD

• Can generate a decent lorica segmentata and the corresponding helmet

• Also has a good grasp of the tunic, and should be able to user others colors than red

• Should work fine with both photos and illustrations

THE BAD

• It's SD: don't expect the leather straps to make much sense

• There's a total of zero images of women in the dataset (not my fault!) so it tends to get imaginative with them

• The cingulum (belt with metal fittings) is partially hidden in many of the photos, so it's a bit iffy

THE UGLY

• The dataset includes a lot of images with pilum (tagged as spear), gladius (sword) and scutum (shield) ...but we know what happens with SD and weapons. Enough rolls and you may get something, but better to go with ControlNet