Sign In

War That Never Was - LoRA (Twilight 2000 4e Style)

Download

1 variant available

SafeTensor

12.73 MB

Verified:

Type
LoRA
Stats

280

Reviews
Published

Jun 7, 2023

Base Model

SD 1.5

Training
Steps: 2,400
Epochs: 16
Hash
AutoV2
B2367C66DE
Trigger Words
tw2k4estyle
drawing
default creator card background decoration
Followers - 138

138

Likes - 2152

2.2K

Donator Badge
04270-3132448077-tw2k4estyle, an nuclear bomb mushroom cloud erupting over a ruined ashen wasteland with an american soldier looking up towards i.png

Hey folks,


Update 6/9/23:
I ended up revisiting this LoRA, and there are some pretty big changes in the new version. I ended up switching to Clip Skip 2 (i.e., trained on animefull). Overall, the style transfer is more faithful, but it does mean that the LoRA now works with a completely different set of models. In addition, I wasn't able to downsize the model and retain full capability, so the file size is larger now. The old version might have slightly better noise offset behavior as well, so think of this as a "side-grade" as opposed to a pure upgrade. Use whichever version you like and let me know if you have any feedback.


Original post:

This LoRA is to replicate the style of the new edition of Twilight 2000, a TRPG of survival in the immediate wake of World War III going hot. The new edition is published by Free League, and this LoRA is based on the illustrations by Niklas Brandt.

Overall, I'm okay with the quality of this LoRA, it's a western comic book inspired lo-fi drawing style, so there's a limit to the crispness/fidelity that I can achieve while remaining true to the source material. I probably won't be working on this one further, but I'm content enough to release what I have.

Usage guide:

  • Use CLIP SKIP 1.

  • Weights from 0.6 - 0.9 should work fine.

  • The art style has a heavy splash art/coffee stain background/edge effect. If you don't want that in the work, then add "simple background" to the negative prompt. A weight of 1.2+ may be required.

  • Due to the limitations of the dataset and Stable Diffusion more generally, getting exact weapons or other military equipment is going to require either a lot of luck or using external img2img or controlnet.