3/24/2024 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. It gets her outfit and helmet a lot more consistently, but sometimes forgets the tiny metal belt around her waist. Hope you enjoy.
Older Description:
The Huntress! She's got more crowbars than Gordon Freeman...
This is a lora trained on 131 hand picked images of the Huntress from Risk of Rain. It was mostly trained on digital art, but there is a smidgeon of 3d art also. It gets the helmet pretty good, but sometimes forgets an antenna, and will put two light things on the chestplate. 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 all of the example images.
NOTES:
-She has 3 little inconsistent bits: Her chest light thing, her antennas, and her scarf. Usually they all behave pretty well, but it can be a little odd at lower weights. Her antenna is probably more consistent if you prompt it (I forgot to do this on ALL my example images :P)
-Despite everything, it can't generate her bow pretty much at all. I might train a "holding bow" lora in the future...
Example prompt: (See example images, all images should still have their metadata. Just drag the images onto the PNG info tab and import it.)
Check out my Artificer model here: https://civarchive.com/models/20139/artificer-risk-of-rain
Description
Updated for use on PDXL! Same dataset, except I changed the tags from danbooru-style to e621-style, since those tags are far more robust. PonyDiffusion's improved resolution, combined with the amount of steps I trained this at, has made her outfit a lot more consistent, especially the helmet and chest plate. Hope you enjoy!
FAQ
Comments (5)
Yeeaahhh! Thanks my dude!
too many training steps, every image i generate looks like its covered in vaseline
Weird, never got a complaint like that before. My generations, and other people's seem fine... Wanna share your prompt or something? I wanna see what you mean
@ZoochMan sure thing, i used the following prompt:
score_9, score_8_up, score_7_up, female, solo, portrait,
<lora:huntress_pdxl_:1>, huntress, huntress \(risk of rain\), humanoid, mammal, robot, robot humanoid, standing, helmet, bodysuit, scarf, armor, gloves, breastplate, unusual eyes, antennae, glowing,
breasts, thick thighs,
standing,
NEGATIVES:
holding object, food, codpiece,
scale 7, steps 20, sampler DPM++ 2M, clip skip 2
I just wasnt happy with the way it looked, and I feel like I had to overspecify to at least make it look like huntress
@Zoey_Zoologist After some experimenting, I'm not perfectly sure why you're prompts are looking like that. I'm thinking it's because of the sampler, I really recommend using Euler A when using Pony, or Restart. Bumping it up to 25 steps might help, too. This lora trained on Ponydiffusion V6, but I suggest using the Autismmix model. It's what I mostly use nowadays.
Also, I tried your prompt with a different negative prompt and was getting better results. Considering trying one of my example images as a base prompt, and build off it from there.
I uploaded an image with a tweaked negative prompt along with more score tags and it looks much better.
Also, I will probably retrain this model at some point, it IS a bit overtrained... and it often gets the amount 'lights' on her helmet or chest plate incorrect (like in my image I uploaded, I couldn't care to cherry pick :P)
Hope this helps 🐛
Details
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.