- Status
Are you using proper negatives? Looks like something easynegative-v2 can fix. Also, the original image has a foot taking up half the screen, the end result if replacing the original image in game would make no sense.
Yeah exactly. The results are fappable at a high denoise level as I predicted, but not true to the original. So if you were to do this to all the ingame images, you would worsen the quality of the game overall. Aside from the fact that every batch processed image of the same character would look like a different person than the previous since we're not editing the result nor using Loras.
Also I'm using an XL model, Hasaku Hentai XL to be precise. I used easynegativev2 before with my SD 1.5 based models, but I've been liking the results, comprehension, and quality of the model over anything SD 1.5 had to offer. It's far superior in every way except for its generation time and compatibility with most loras and embeddings, since they were created for SD 1.5 based models.
Last edited:
It's almost 1july , hopefully fvlaenix is doing fine with debugging
It's all about work flow. For example, plug in your original image into control net, use lineart anime, and produce something like image 1
Then either plug in booroo tags, or make your own custom prompt for each image -
You don't have permission to view the spoiler content.
Log in or register now.
use a decent model, play with the settings to find what has good results for both input and outputs and pump out stomething like this -
Not much more work needed, basically coloring, maybe some lighting and detail tweaks, but you get the idea... It's a workflow. Establish it in a way that works for a couple radically different images from the same workgroup to get consistent results then batch run.
Edit: You can see multiple issues, like the lack of string bra, and other details that are off. Booroo tags would likely make a huge positive difference, and you can generate them in batches for given images.
Last edited:
- May 25, 2022
- 55
- 175
Man what a trip thank you very much ! It sure looks complicated, but the pictures are hella funny, the last one sure looks good although not bearing the original character identity as you said, seems to me that it's harder to work with details such as a smirking face, peculiar expressions like her eyes being mildly closed, the grins and everything else, its true lot of TOT faces and art are very expressive, hard to simulate with an AI that can go out of control at any time, would probably ask hours of work to get a proper result similar to the org renders
I don't know the ai tools at all but I do hope that some devs will end up using it, not to do all the work on itself but a proper mix, I wonder if Bo would find an use for it with his games
Does anyone have a completed save file? I've completed the game but accidentally deleted the whole game and I kinda dont want to waste this much time again just to replay some scenes. Pls
Alright, finally got around to making Yvonne's artwork as per your request and not batch prompted lol.
Enjoy~
Edit:
Generation Details + Image post on Civit.ai -
You must be registered to see the links
Last edited:
Is there any way to reupload this photo eslewhere, I can't watch related files on f95zone.to I don't know why
- May 25, 2022
- 55
- 175
Fantastic one much like the org ! lol'd at the squid game jacket and glad you managed to save her bra too, one of the cool things about that maiden
Jessica is one of those strange cases where she's kinda tough bcz her outfit is so non-standard. She's got that entire skintight bodysuit with one side black and one side white, plus the hat. Ill try it next tho lolYou know who actually needs AI art? Jessica
- Status