thanks, playing this for the NTR
In EP II, there is a big part where Taylor goes to the beach alone and ends up in several NTR situations. Also in EP. II there is a part where the archivist, Feng Min, does NTR stuff with denizens of the virtual world. That update is available in my patreon, and will be available here in late december.
I really liked the game, I didn't think such a consistent game could be created with artificial intelligence, really congratulations, I hope it gets better.
I am also interested in these issues, I have a humble question: Is 10GB VRAM (3080) enough to use models that produce images of this quality? Or can I not run the model at all?
I think in a 3080 you can run most of the models I use. The trick is to find a model that you feel confortable working with, and don't underestimate the importance of knowing how to composite in photoshop/GIMP. AI is never really good enough alone. That's why so many people complain and immediately shun an AI game. I think of AI as an oven for baking bread. After the bread is done, you need to find good cheese, ripe tomatoes and tasty onions to put inside. Compositing, lighting, composition and layering are those extra ingredients to make it really appealing and tasty.
Thanks for a detailed response. Does SD allow NFSW renders? Some said SD only gives non NSFW renders. Is there any special code you had to put in for that? And any comment on Unstable diffusion?
Stable Diffusion is just the technology, I'm not talking about any web based service, although there are quite a few. SD runs locally in your computer, using your graphics card, and as such is completely uncensored, if the model is. There are some better than others, in my case I use several different PDXL cooked with different LORAs and custom VAEs to get my results.
On web based services, I sometimes use CIVITAI and there are a few others that let you generate NSFW art, but CIVITAI is by far the biggest. The amount of models, articles and community support makes it possible for people on the road or without powerful PCs to create beautiful illustrations online. Never heard of unstable diffusion.
SD has many NSFW options (although there are various version of SD, each with a varied level of inbuilt censorship). A lot of people use PonyXL for NSFW as it has a lot of booru training data.
Those are my goto models. They have a steep learning curve, but once you get your formula correct, you can almost generate anything you need, given you have time and patience to sit and sort trough hundred of images to find 5 or four that are usable with a bit of post work. Sometimes I like a face but not the hands or the light, and that's where Photoshop comes in, combining two or sometimes even three or four images into one using good ol photobashing techniques.