Type | |
Stats | 372 3,463 1.3k |
Reviews | (36) |
Published | Jul 29, 2024 |
Base Model | |
Training | Steps: 3,200 Epochs: 2 |
Usage Tips | Clip Skip: 1 Strength: 1 |
Trigger Words | roguetraderstyle |
Hash | AutoV2 6D3A5FBBB2 |
Trained on the character portraits from 2023's Warhammer 40K: Rogue Trader by Owlcat Games, this LoRA should help generate custom character portraits in the style of the game's existing art. Note that alien and abhuman character art (such as Eldar, techpriests, Astartes, and the Navigator) were excluded from the training dataset, so this is pretty focused on generating player character art. Focused on style and general aesthetics; this LoRA won't know specific 40K concepts any better than the model you're using.
Trigger word is "roguetraderstyle". Starting the prompt with "roguetraderstyle full length portrait of..." works consistently.
I recommend starting at half resolution (540x740), and then switching to the game's expected portrait resolution (1080x1480) with hires fix. This helps reduce having too much "greeble", since 40K characters already tend to have a lot of drip.
Along with the LoRA itself, samples were generated using Demon Core checkpoint and the Negative Dynamics XL embedding. Unedited, so what you see is "out of the box" and could be vastly improved with inpainting, controlnet, etc...