Duke4.net Forums: Adventures in Stable Duke-Fusion - Duke4.net Forums

Jump to content

Page 1 of 1
  • You cannot start a new topic
  • You cannot reply to this topic

Adventures in Stable Duke-Fusion  "Let AI do the art"

User is offline   Rellik 


Rather than cluttering up the ERSGAN thread in general gaming, I figured I would document my attempt at re-imagining the babe posters (and anything else that fits) with Stable Diffusion.

Frankly, I am surprised the rapid pace that AI art is advancing now. As long as you have a decent graphics card and a decent amount of memory, it's easy to get into.

Software needed:

  • Automatic1111 WebUI
  • The correct version of python (3.10.6 as of now)
  • At least 1 ai model, one might be downloaded when you first install, but you can get others as civit.ai, Deliberate is a popular choice
  • (optional, but helps with some finite control) controlnet and its associated models, installable within the webui, models here

So far I have done the two Tonight posters as seen in the ERSGAN thread

Now my next challenge is Sister Act III, and this was a fun one.

Initially I had trouble getting img2img to pick up on the poses, it's also a pain in the ass to get the correct clothing and hair on two models, thankfully both problems can be solved with minimal extra work.

This is where controlnet(pose) and inpainting comes in.

First, controlnet has a pose model which will detect the pose based on the models position and give a fairly consistent output, I could not get the left model to be as close as I like, but I think it's fine with the 'story being told'
Attached Image: tmpn10qx8jf.png
Funny thing about pose is you can get quite a wide variety of styles, but with the same pose, its kinda cool. And frankly I don't know what style to settle on, realistic-ish or cartoony? There is so much variety
Attached Image: 00151-1365475421.pngAttached Image: 00006-1190816479.png

For the moment I have done the realistic-ish since it goes with my previous two

Attached Image: sa3v4.png

There are some differences that I was able to use inpainting to blot out undesired aspects and re-roll something new by editing the prompt slightly.

User is online   Ninety-Six 


Realism makes the most sense I think, since the art style used to draw them is identical to the way "real" women are presented in-game. Most importantly being the two girls themselves, looking pretty much identical to the poster when you run into them in XXX-Stacy.


Uncanny faces

User is offline   Phredreeke 


I used Level 4 to batch process a bunch of Duke tiles.

Here are the prompts

tile0963: realistic, a woman in a red bikini posing for a picture, 1girl, solo, red panties, red bra, breasts, blue eyes, english text, underwear, panties, lying, rating:safe, abs, lips, medium breasts, realistic, bra, red bikini, armpits, long hair, fake cover, cleavage, digital art, by Hugh Hughes, sots art, torchlit, package cover, in clothes! highly detailed, hot coffee, high quality, masterpiece
(images 60-104 has tonight hot hot hot added as an additional tag at the end)
tile0577: realistic, a woman in a bikini standing on a beach, a raytraced image, inspired by Boris Vallejo, ocean, beach, 1girl, rating:safe, swimsuit, cloud, solo, horizon, barefoot, bikini, water, white bikini, sky, cloudy sky, day, long hair, waves, outdoors, sand, breasts, black hair, sun, navel, shore, standing, style of mirror\'s edge, [ rigidly defined abs ]!!, leg shot, high quality, masterpiece
tile0965: realistic, a woman in a bikini and shorts sitting on the ground, posing with rocket launcher, inspired by Tina Blondell, 1girl, breasts, solo, rating:safe, english text, sunglasses, blonde hair, cleavage, photo \(medium\), fake cover, large breasts, barefoot, red bikini, blue shorts, realistic, movie poster, now playing attack of the bleached blonde biker bimbos, high quality, masterpiece
tile0766: realistic, a woman in a bikini posing for a picture, a poster, inspired by Torii Kiyonobu I, cobra, 1girl, swimsuit, bikini, nail polish, breasts, red nails, red bikini, solo, rating:safe, photo \(medium\), toned, realistic, navel, ocean, long hair, makeup, lips, lipstick, hair ornament, looking at viewer, lena oxton, enter night, high quality, masterpiece

(if the prompts don't make much sense it's because I used CLIP interrogator results from the HRP as a starting point, and then added tags from deepdanbooru, and also added some custom edits.

and the negative prompt (same used for all images): bad result, worst, random, invalid, inaccurate, imperfect, blurry, deformed, disfigured, mutation, mutated, ugly, out of focus, bad anatomy, text, error, extra digit, fewer digits, worst quality, low quality, normal quality, noise, jpeg artifact, compression artifact, signature, watermark, username, logo, low resolution, worst resolution, bad resolution, normal resolution, bad detail, bad details, bad lighting, bad shadow, bad shading, bad background, worst background.

The images were processed with 30 steps and an AI strength of 0.35, 0.5, and 0.65.

Attached thumbnail(s)

  • Attached Image: 00115-11.jpg
  • Attached Image: 00154-5.jpg
  • Attached Image: 00075-1.jpg
  • Attached Image: 00083-9.jpg
  • Attached Image: 00246-22.jpg
  • Attached Image: 00266-12.jpg
  • Attached Image: 00316-17.jpg

Attached File(s)

  • Attached File  jpeg.zip (6.72MB)
    Number of downloads: 32

This post has been edited by Phredreeke: 02 April 2023 - 05:24 AM


User is offline   Phredreeke 


Last night I again processed a bunch of tiles using a new model

Prompts should be mostly the same as I used for Level4 above, with the addition of these prompts for tile0964:

realistic, a couple of women standing next to each other, multiple girls, 2girls, realistic, magazine cover, rating:safe, bra, underwear, breasts, fake cover, panties, looking at viewer, medium breasts, black hair, navel, red hair, cover, underwear only, photo \(medium\), arms up, concept art, inspired by the Brothers Hildebrandt, feminist art, character art, high quality, masterpiece

For 75 and beyond in lofi-jpeg2.zip and the entirety of lofi-jpeg3.zip I had blasted the input image with noise beforehand and then added noise, noise, noise, noise, noise, noise to the negative prompt.

Attached thumbnail(s)

  • Attached Image: 00040-26.jpg
  • Attached Image: 00052-8.jpg
  • Attached Image: 00061-17.jpg
  • Attached Image: 00212-17.jpg
  • Attached Image: 00206-12.jpg
  • Attached Image: 00213-19.jpg
  • Attached Image: 00219-25.jpg
  • Attached Image: 00147-28.jpg
  • Attached Image: 00143-24.jpg

Attached File(s)


User is offline   MrFlibble 


Cool, but this model (or are these two different ones?) seem(s) to really struggle with faces, and in some cases, body proportions too.

User is offline   Phredreeke 


Posted Image

Used the LazyMix model for this with an image of Bombshell that mark posted as input

User is offline   xMobilemux 


Would any talented people in AI art be willing to try and recreate and upscale the background menu from Duke Nukem Land of the Babes?

This one:

I've wanted a high resolution wallpaper of that background ever since I first saw it and my attempts at upscaling it often come out looking worse.

User is offline   Mark 


Any idea what that backround is supposed to be? I brightened it up and couldn't tell.

There must be a high res version of that out there somewhere. I found it and made a movie poster out of it for the HHR mod.

Attached thumbnail(s)

  • Attached Image: 3391.jpg

This post has been edited by Mark: Today, 05:59 PM


Share this topic:

Page 1 of 1
  • You cannot start a new topic
  • You cannot reply to this topic

All copyrights and trademarks not owned by Voidpoint, LLC are the sole property of their respective owners. Play Ion Fury! ;) © Voidpoint, LLC

Enter your sign in name and password

Sign in options