Sign In

Celty Sturluson - Durarara!! [PDXL Update]

522
2.4k
6.7k
54
Verified:
SafeTensor
Type
LoRA
Stats
464
6,669
Reviews
Published
Apr 27, 2024
Base Model
Pony
Training
Steps: 10,000
Usage Tips
Clip Skip: 2
Strength: 1
Trigger Words
celty
celty sturluson
Hash
AutoV2
BFA8E95A53
default creator card background decoration
ZoochMan's Avatar
ZoochMan

4/27/2024 PDXL Update:

I updated this model for use with PDXL. Please see the "About this version" section to the right for details. It's pretty much the same model, just trained on better tags, at a better resolution, and with a better optimizer on a better model (PonyDiffusionV6). Hope you enjoy!

-The little weird symbol on her helmet is still pretty inconsistent, but doable if you use inpainting (if you want it to be accurate)

-There is a weirdly HUGE bias towards there being a motorcycle in every generation, so moderate effort is needed to not have one. Consider only including "helmet" in the prompt, and nothing else like "bikesuit" or "motorcycle helmet". If you give it an inch, it'll take a mile. Prompting for indoor environments help, too.
-The back of her helmet is supposed to be blue, but since there isn't very much art of that detail, it might not get it very well on rear-views. Also, she seems to always want to look forward, or at the viewer, meaning you'll probably not see the back of her helmet unless you really prompt for it.

Fun Fact: I had to train this model 3 times. First time was a failure since I tried to use a new optimizer in training, but used bad settings which completely bricked the model, and it ruined every generation. And the second time was a failure since I accidentally had "flip augment" enabled, which made good generations, but her odd helmet symbol would appear on BOTH sides...

Older Description:


Celty Sturluson from the anime Durarara!! - Full disclosure, I've never seen this anime, but I loved her design, and... here we are!

This is a lora trained on 87 hand picked images (and 1 gif by accident lol) of Celty from Durarara!! It was exclusively trained on digital art, no 3d content. It gets the helmet pretty good along with her suit, mostly because her design is pretty simple besides the helmet. It was trained on danbooru-style tags, so use those when trying to prompt - I've included the tags in a .txt file in the "Training Images" download below.

It works pretty well with most ckpts, but I recommend Anything 4.5, I used it for most of the example images.

NOTES:

-The helmet could be removed if you prompt no helmet and define the facial features, and lower the weight... But why would you do that?!

-It will still have issues generating vehicles/motorbikes, but it can still kind of work on some models if you're lucky

-It can generate a "male" version of this character with the right prompting

Example prompt: (See example images, all images should still have their metadata. Just drag the images onto the PNG info tab and import it.)