Using Lora AI models in VAM

[chronic82]

Hi everyone,

I lack the technical nous to achieve this, but I'm hoping I can inspire someone who's already up to speed with a lot of the technical skills and knowledge required to make this happen - if it is possible?

I can't remember which service enables this as I stumbled across it about 6 months ago before I'd even heard of VAM, but I think it has something to do with the Flux suite of generative AI products. I believe it is called Lora. You upload as many different pictures of face angles and emotions by the same person as possible, then you can use generative AI to put that person in any scene with lifelike photorealism. In general, you want pictures of all angles of your head and all main emotions. This would enable us to put ourselves in any scene in theory (people already us this tech to put themselves in boxing matches against Tyson, for instance), or put anyone else who allows us enough of their pictures, and it would like exactly like that person from any angle.

I'm assuming also that plenty of AI image gen software could be used to generate a "master" face if you like, a straight face pointing straight at the camera, then use that as the template to tell the image gen software to make a side profile, a smile, a frown etc, then we could create fully fictional lifelike humans with consistent appearances across all emotions and situations. Looking through the forums, it seems people want only a straight face looking forward to use to make their dream girl, but surely this returns less than impressive results, because only a straight face pose in VAM would look like the person in question, whereas VAM would have to guess what the person's smile, frown or side profile would look like?

Really hoping I've inspired someone to look into this who can report back with some exciting progress!

 

[LubaFoto]

Well, the thing have already been test Ai and Facegen, bu it's scanning 2d pictures not 3d, so the result is at best ok. The problems is texture ai is not very good at it and so is for facegen. Second, for expression, vam, Daz or even AI, have no way to know bones and muscles structure, so for now to make a realistic character, a creator will have to create first a realistic shape morphs of a person and then create uniques poses morphs.
Maybe in the future?

 

[chronic82]

Well, the thing have already been test Ai and Facegen, bu it's scanning 2d pictures not 3d, so the result is at best ok. The problems is texture ai is not very good at it and so is for facegen. Second, for expression, vam, Daz or even AI, have no way to know bones and muscles structure, so for now to make a realistic character, a creator will have to create first a realistic shape morphs of a person and then create uniques poses morphs.
Maybe in the future?

But is there not a way we could import these lora files into the game? The body isn't reliable, but we could build the body with consistent shape and textures in VAM, and without face detail its easier to get ultrareal looking bodies, but this program enables you to reliably produce an avatar with a 3d head and consistent expressions all the time. We'd in theory just need to import the head

 

[LubaFoto]

Yes you could, Lora is 2d generated, then you send the result images to Facegen (facegen will create a 3d daz genesis model with it) and Facegen to Vam. Unless you have higth quality image from Lora, it won't be good. My personal thought is the thing does not work very well.

You could ask the forum, about Facegen or an alternative, maybe somebody had better result.

 

[chronic82]

cool thanks, worth looking into