FAQ needs more Q's
Okay - I think the FAQ needs work. Here some info I think is missing...
1. How was Daz AI trained? i.e., using their own sources (like Adobe's Firefly) or just taking other people's work (like Stable Diffusion)?
2. In regards to this quote in the FAQ - does this mean you're going to be selling AI-only characters in the future? Or you 'unlock' the character if you own, say, Jennie 9?
Our character products are pulled from Daz 3D’s large library of content to help direct your creative flow. These products mimic the specific style/look of the various Characters available in the Daz Store giving you a much more predictable outcome. Our Free and Paid users will have access to our core characters Victoria 9 and Michael 9 with other characters available for purchase from the store in the future that you can add to your account.
3. Who owns the rights to the image? (question #1 might influence this).
Comments
Question 1 is literally the only one I personally care about, and influences whether I'm doing any further business with DAZ, and it is troubling that it isn't clearly answered in the FAQs/product page already since the team knows how much of an issue this is with other generative tools.
Hoping to see this information updated quickly.
I'll add, for starters:
"What are the content restrictions on what you can produce?"
"What are the Legal TOS?"
"Can you upload your own models?"
"Can you alter models, say with Zbrush?"
"Who owns the rights to the images created?"
How can anyone own the images when they cannot be copyrighted?
It looks like this is a new product to pull in users who don't already use DS and aren't up for the learning curve, not something that was ever intended to make existing DS users happy.
For us DS users, we still have our program and our libraries are still good to go, so even though this product does not yet seem compelling to me, I don't think there's any reason to panic or get angry about it.
If using our existing content is not an option, then the FAQ needs to clearly state this. If this is the case, then AI Studio for me is not worth it. I'm not going to pay for AI only stuff when I have an extensive library in which I've already invested and will continue to do so. The extra 3.99 was supposed to include 'characters' but I don't see anything other than M9 and V9.
Yes, this is an extremely important question. Did they just use the galleries on the site as training data? I'm pretty sure no asking for permission happened.
Another question: How can one opt out from getting into the training data (beside never using DAZ gallery again)?
Making it opt in (with not being opted in default) would be ideal but I doubt they'd allow that.
It has been answereed in the anouncement thread in the Commons: they trained it using Daz-owned images.
https://www.daz3d.com/forums/discussion/comment/8658501/#Comment_8658501
From what we've been told Daz did "not" use the galleries to train.
"Who owns the rights to the images created?"
(Edited by mod to pull the questioned text from the quote)
You do - the EULA has been updated to relfect this (but note that, as far as we are aware, it is not possible to copyright a bare AI image - it needs some human editing too).
So, if I use the AI generator to make a background like with a castle I can then use it like I would any regular render I would make? I own like 4 gigs of stuff, but I hate setting up backgrounds. Also, I can't seem to find the EULA anymore....can somebody tell me what page it is on, please?
Yes, that would be allowed.
https://www.daz3d.com/daz-licenses
I'd like to how seeds work with regards to AI art.
All the AI geernators use a pseudo-random process to gernate a scattering of data, then iteratively refine that to bring it into accord with the range of values triggered by the prompt. The seed value determines the starting point for the pseudo random number geenration, so in principle the same seed, prompt, and training data should always yield the same result. I'm not sure if that is quite how it is implemented though, as the instructions seem to say using the same seed can generate subtle variations (my attempts gave the same result each time, but another mod got some differences)
If you keep the seed (which you can add by clicking the little icon on the image, next to the save button) and edit the prompt you will get a variation of the image rather thana completely fresh version - so you can try to refine something that was close. Radical alteration of the prompt will probably break that, however, though I was in my test able to go from explorer hacking through jungle, misty, machete to explorer using machete to hack through jungle, misty without greatly chnaging the result (other than getting him to actually use a cuttin implement, which looked more like a sword than a machete). Sorry, I closed the window so I can't show the results of those prompts.
I took pictures of belly dancers with Daz AI. Some of the costumes and jewellery were so exquisite that I would buy them immediately for my 'real' DS renders if they were available here in the shop. Will that be possible?
That would be great, if some brave PA artist would provide a tools and tutorial
how to go from image to 3D objects, like for example from clothes images to dForce ready Daz Studio cloth.
The text to 3D tool, that was announced last year, is being worked on and will be a part of a future update.
Thanks for the info about future updates, @frank0314