My models are just a mix of everything that you yourself can find on this wonderful site. I'm not chasing fame, I don't need your money, I just sometimes spend a few hours of my free time experimenting with my favorite models. The page with this model exists on this site only because once my friends asked me to share their mix. And because it's just damn convenient. Please don't accuse me of stealing someone's property. Considering what we're all doing here, that's a little ironic. In my mixes, I use such checkpoints as Luma, RevAnim, DDosmix, Perfect World, GalenaRedux, DarkSushi, CosmicBabes and so on. I don't even remember all the model names. I don't do anything that you can't do yourself.
Try, experiment, enjoy!
Here is the workflow for my pictures:
Sampler: DPM++ 2M Karras (because it runs faster on my weaker hardware, while giving a good result)
Base image size: 512x768 (because I don't have enough VRAM for more).
CFG Scale - 7. Steps - 30.
Default negative line: EasyNegative, drawn by bad-artist, sketch by bad-artist-anime, (bad_prompt:0.8), (artist name, signature, watermark:1.4), (ugly:1.2), (worst quality, poor details:1.4), bad-hands-5, badhandv4, blurry, child, loli, kids
Then I generate 5 pictures to identify any obvious problems, after that I make another 10-20 pictures, choose the best one from them and proceed to the upscaling stage.
I turn on the option Hires.fix
Upscale by - 2 (because I don't have enough VRAM for more)
Upscaler - 4xUltraSharp
Haires steps - 15 (half of the steps from the first stage)
Denoising strength - from 0.3 to 0.6 (the larger this value, the more detailed the picture will be, but the greater the risk of getting artifacts and distortions in those places that the neural network considers doubtful or incomprehensible)
Turn on ADetailer. The 1st is - face_yolov8.pt, the 2nd is the mediapipe_face_mesh_eyes_only. If necessary, I fill in its positive and negative lines, but more often this is simply not necessary.
PROFIT!
Description
FAQ
Comments (47)
So... what's new in v7b?!!?
+1 lol what gives?
Oh, the release of this version happened before I had time to prepare for it! In general, in this version it became possible to generate dickgirls not only with balls, but also with pussies (without testiqles). I hope that in this case the flexibility that I so value in my model is not greatly affected.
@Iris_DS so it's kinda like futasun but a bit more flexible?
@Malessar Yes
is there any image quality difference between this version and the last one? or is the only difference the plumbing options on the women?
Hello, could I use version 2.0 to create a mix? If not, no problem. You do a great job :)
Still a good model but I feel like it's not as good as v6. Character's face looks way older. Some Lora work great in v6 don't work well with v7. I hope v8 will be an improved version from v6 instead of v7.
rich model
A smaller file simply takes up less space and is easier to download. This is achieved at the cost of a slight reduction in image quality.
Does anyone have any tips for generating African American girls/tgirls with DarkSun? I've tried prompting with "african american", "dark skin", "very dark skin", "african", and none of them give me much darker than a mocha-skinned girl. They're pretty and all, but they're not much different from a tan version of what I'd get without those prompts. I'm talking actual dark skin - that warm chocolatey brown or even darker. Like from Oprah to Wesley Snipes or Lance Reddick.
Try using people mixing. Add something like [[Oprah | Lance Reddick | Wesley Snipes]:0.1] to your prompt. If it works, try playing with names and the number. Less means more influence. You may see more related examples in my prompts.
@Olbanets Thanks! I'll give it a shot.
@ttb2112 let me now how it works for you
@Olbanets no luck. It didn't seem to make any difference at all. I even ran an X/Y plot with increasing weight all the way up to +1.0 - no change in complexion - very slight change to posture/layout of the image. <shrug>? When I gave up on that approach, I threw in an embedding of a black girl. Shockingly, that didn't seem to make a difference either. I must have something squirrely in my prompt or settings.
I had some level of success using "dark skin" and "african woman", but the model doesn't do very dark skins.
I had this same problem. Use a LoRA that's been trained on the concept of dark skin like https://civitai.com/models/114568/dark-skin-tan-control and it should work. Character LoRAs that have been trained on darker skin will work too, you can look up the correct tab by opening the LoRA metadata.
You could also use a different model and then apply DarkSun using img2img.
https://civitai.com/posts/4896754
is she black enough? ;-)
@Olbanets blast from the past! LOL, no, she's too black. I think you left her in too long.
This is my favorite model on the entire site. Come back and bless us with more awesome AI content
my gens are coming super anime-y instead of the semi realistic style of darksun, idk why, it isn't any loras... any advice?
Just add photorealistic or photorealism on the prompt, you can also put anime on negative...
Same here. Prompts don't change anything. Very frustrating.
@Murdochella something isnt working like it used to
Anyone have any recommendations for an SDXL model that is similar to DarkSun: tgirl/futa friendly, 2.5D, and produces great results? I've been messing with SDXL lately and really struggling to get my favorite style back... seems like most of the 2.5D SDXL models lose coherence as soon as you add tgirl prompting. :(
I looove the style here. Unfortunately my images are always distorted, not with artefacts or so, but weird bodys with multiple arms or twisted legs and such... Working on it, though.
It is one of my favorites, I would greatly appreciate it if you continued perfecting it.
How to use this only in FaceDetailer step? First loaded model is PonyXL. It is giving me error (probably because the start model is SDXL and the other is SD).
You can always use the pony generated image to start FaceDetailer with any SD1.5 model, can't remember the pack I used, but in some cases is through latent, others using controlnet reference, it depends.
This model is insane. The quality, level of details, and good look on every character created. Love this. keep up the good work. TY.
Not this Anime/hentai shit.
I think this guy was actually born yesterday. Welcome!
Ever considered bringing this over to pony to take advantage of the increased prompt flexibility that pony offers?
A magnificent semi-realistic model! I really love this magical mysterious dark style. All images are generated very expressive. And of course, I am pleased with the excellent flexibility and responsiveness. The perfect job!
Besides the semi-realistic scenes DS renders, it has a bunch of built in prompts that other models require loras to render. There are several bondage or tie-up prompts it will execute, and execute well w/o a lora. It has a very wide repertoire of attire, especially fetish attire.
Wow- this model is superb. I just started using the pruned model today and am really enjoying it.
Absolutely astounding, loving this model so much!
i wish there were a pony version
One of the few models that can generate dark images directly out of the box. Too bad there's no PDXL version.
This was my FAV model...i came back after a year or so and im sad that it hasnt been updated, it can produce fantastic images without loras. Very underrated
Came back to PROOMPTING after few years. Switched from automatic1111 to comfyui and I just can't seem to get anywhere close to my old automatic1111 generations.
Rest in peace my favorite model.
Details
Files
Available On (2 platforms)
Same model published on other platforms. May have additional downloads or version variants.