Has anyone combined DAZ3d with AI rendering? Or using DAZ to create custom characters?

SasquatchIsCool

I've been messing with AI a bit and I think it's biggest strength is when coupled with AI.  I've used DAZ3d to create various characters and then put them in scenes.  Have any of ya'll tried this?  Can you share your daz3d AI combinations if you have?  Is it something you'd like to learn?  

WendyLuvsCatz

https://www.daz3d.com/forums/discussion/591121/remixing-your-art-with-ai#latest

Matt_Castle

You should probably join up with the folks in this thread: https://www.daz3d.com/forums/discussion/591121/remixing-your-art-with-ai/p1

SasquatchIsCool

Matt_Castle said:

You should probably join up with the folks in this thread: https://www.daz3d.com/forums/discussion/591121/remixing-your-art-with-ai/p1

That seems really cool.  I don't see anything as new as I'd thought.  I thought for daz it would have been a big deal.   

Oh here is a DAZ3d to AI to Videos https://youtu.be/_YZibARM3XQ

Matt_Castle

SasquatchIsCool said:

I thought for daz it would have been a big deal.   

Some people love it. Others don't.

Even if we put aside the questions of copyright and IP (which put off many), algorithmically generated images are not completely consistent, struggle in a lot of cases, and often have strong biases.

Running an image through an AI at a high enough strength to successfully change the skin or hair to look more realistic can significantly change a character's appearance or expression, can alter the lighting direction, make once strong and distinct shadows disappear, etc. I've also found it's generally very poor at depicting faces from side or rear angles, given a predominance of portrait photography in most training data sets. It also often just fails to make sense of what you're feeding in; even with a description, something too far outside the training data is likely to get reinterpreted.

It provides visually impressive results in some cases, but it has issues in many others, so you're giving up consistency and control for something that a lot of the time many people can still tell is artificial (particularly so when talking about a community of trained artists).

Even if it provided perfect results, just layering filters over an image doesn't really add new story telling tools to your arsenal. People are happy to take in scenes depicted in watercolour painting, comic books, 20-year old game engines, etc. DS has never been top of the line graphics anyway (respectable for consumer grade, sure, but certainly not feature film standard) and its users are aware of that, so it's not all that surprising that a lot of the userbase aren't in dogged pursuit of ultimate visual fidelity.

 

 

SasquatchIsCool

Matt_Castle said:

SasquatchIsCool said:

I thought for daz it would have been a big deal.   

Some people love it. Others don't.

Even if we put aside the questions of copyright and IP (which put off many), algorithmically generated images are not completely consistent, struggle in a lot of cases, and often have strong biases.

Running an image through an AI at a high enough strength to successfully change the skin or hair to look more realistic can significantly change a character's appearance or expression, can alter the lighting direction, make once strong and distinct shadows disappear, etc. I've also found it's generally very poor at depicting faces from side or rear angles, given a predominance of portrait photography in most training data sets. It also often just fails to make sense of what you're feeding in; even with a description, something too far outside the training data is likely to get reinterpreted.

It provides visually impressive results in some cases, but it has issues in many others, so you're giving up consistency and control for something that a lot of the time many people can still tell is artificial (particularly so when talking about a community of trained artists).

Even if it provided perfect results, just layering filters over an image doesn't really add new story telling tools to your arsenal. People are happy to take in scenes depicted in watercolour painting, comic books, 20-year old game engines, etc. DS has never been top of the line graphics anyway (respectable for consumer grade, sure, but certainly not feature film standard) and its users are aware of that, so it's not all that surprising that a lot of the userbase aren't in dogged pursuit of ultimate visual fidelity.

 

 

Here is where I think it would excel, though especially as the tools progress. DAZ3d would allow to make characters.  Very specific characters.  You can then upgrade them and put them in a story with full animation.  

I'm working on that process now.  Because control and consistency are becoming a thing and the thing that allows for that is having video or images as the base pushing it forward.   And if we can get consistency and an almost 1:1 like rendering then we would be a new area.  

I'm already working on that but intersted if people would like that or not to learn. 

Richard Haseltine

Closing this therad since there is already at least one on the topic.