Hello BadIdea4, I’m currently looking into that. I find that the faces you mention are used so much, because they just work. A lot of my other experiments have resulted in malformed faces.
I have found something that creates a more unique style, namely increasing the generation resolution and the Add More Details lora. This has overall resulted in a higher fidelity, so it leans more towards the semi-realistic style. The face does look very nice in my opinion. What do you think?
I like it! I think the black background makes the face look a little flat / washed out, but that probably won’t be a problem against a real background.
That settles it then. I will go for something in-between semi-realistic and anime, like posted above. I would like to put in the extra work, since nothing worthwhile is easy. My thanks go out to everyone for voting!
That’s a shame. According to the poll most people would agree with you, but due to the fact that it’s 60% vs 30% I prefer to take the middle ground. Also partially because of personal preference
Blending styles might not be too difficult to do if your workflow uses a mixture of realism and anime-influenced models/loras, I’ve seen it done here and there with varying success. Of course, adding weight gain on top of all that makes things a lot more complicated.
I also definitely also prefer the first look to this one, though I always default to what the creator wants because that usually ends up making the highest quality work.
I prefer the more anime style because if you go realistic there is the whole headache of AI randomly getting close enough to a real person that it may cause some issues down the pipeline if that person finds out this exists
@cheddar@jcc309@Manchura
Thank you all for your insight. There are some things you mentioned that I hadn’t considered.
You are correct that weight gain is more difficult to reflect with AI, especially in the more realistic styles. It all usually ends up looking unnatural, or at best just ‘off’. Sticking with the old style would be easier because it would save me from having to start over. I also think it looks nice, so losing the fun in creating is not an issue.
I find that Manchura makes a great point. I don’t want to get in trouble IRL over a fetish game I make in my spare time. Then again, a simple disclaimer at the first startup can do wonders.
I also would like to mention that I have made several attempts at using DAZ 3D in the past, so there’s that. I can already think of a few critiques, including what Manchura said regarding realism. Maybe a poll would do that idea justice. I’ll give it some more thought first, so I don’t make decisions too hastily.
AI doesn’t work by copy-pasting individual inputs, though. To get one to draw a certain person, you either need to prompt for them by name, describe them in detail, or train the model hard on that person. And to the extent that common models are over-trained on a few people, it’s famous people, who most artists would adjust away from, and who can’t really do anything about unauthorized use of their image - it basically belongs to the public anyhow.
