Posted on 2023-10-08 18:19:18
Score: 2 (vote Up)

It looks like Nightfall was caught by the enemy and reprogrammed with a new goal in mind.

Was hoping to get the extra Hilda panel out either today or tomorrow since the milestone was reached, but with Thanksgiving happening I've only gotten it about half done so far. In the meantime, here's hypno-edit of a post from yesterday.

Posted on 2023-10-08 22:03:31
Score: 0 (vote Up)

Very nice: Do you mind showing the workflow for the first txt2img you used :D?

Posted on 2023-10-08 23:10:04
Score: 1 (vote Up)

Oh my gosh that's good~~~ <3 <3 <3

Have a good Holiday Guile!!!

Posted on 2023-10-09 02:04:53
Score: 0 (vote Up)

Wonderful to see her enjoying herself! She'll be happier that way rather than pining after a taken man~

Posted on 2023-10-09 19:47:30
Score: 1 (vote Up)

Sure. As a quick note, posting links in comments is still not fixed so you need to remove the spaces in the urls to see the images (NSFW obviously).

I genned initial image at 512x768 with hires fix enabled at 2x at a denoise of to 0.4 (Anime6B upscaler). I use the default a1111 parameters for everything else (CFG 7, Euler a sampler, 20 steps). This is the prompt I used:

(masterpiece), (best quality), <lora:GuileStyle:1>
BREAK <lora:nightfall:0.85>, nightfall, white hair, hair over one eye, breasts, (expressionless:1.1), serious, completely nude, nude, nipples, collarbone, (medium breasts:1.2), (abs:0.85)
BREAK hetero, vaginal, penis, uncensored, torso grab, cowgirl position, looking at viewer, straddling, arm support
BREAK pov, looking at viewer, indoors, bedroom

I got the pose by modeling it and passing it through the openpose controlnet at a weight of 0.5 to get this: imgbox . com/UU3jkCfn.
The Nightfall lora was one that I custom trained.
This was the initial gen: imgbox . com/RbFHyMqz

Maybe I can write up a more detailed process on the forums at some point since the initial txt2img gen is only the first step.

Posted on 2023-10-10 05:44:12
Score: 0 (vote Up)

Thanks for the reply, Guile; it's always useful for me to see how other creators generate their outputs.

For me, currently, the biggest struggle is to create the sort of slight "jaw-drop" open mouth from the expressionless "expression": it's either always a closed mouth (sort of "._.") or frowning when I type it down. How did you got yours? Adetailer with a different prompt, or a careful work of Inpaint? Tried both, but I guess I'm writing my prompt badly.

Posted on 2023-10-10 05:48:19
Score: 0 (vote Up)

Oh, and almost forgot (damn that I can't edit my comments any more): which checkpoint you used for this one? Meina?

Posted on 2023-10-10 21:31:02
Score: 1 (vote Up)

I tend to redraw all my expressions (mouth/eyes/eyebrows) from scratch after the initial gen. I gained some experience from back when I used to manip other people's works to hypno-fy them. You might be able to train an "expressionless" lora though to help get the expression you want since there's more than enough reference material for it.

As for the model, I use Based65 as its default style meshes well with my style lora. From what I've heard though, Based67 is what you'll probably want to use since it apparently works better with character/concept loras and is easier to prompt with.