Denial

[ADAM]

One of the reasons I still haven't abandoned Daz is I'm using Daz as a baseline and AI to touch it up. That's the only way I've been able to get a step closer to realism.

Daz is as stated more customizable to my needs. AI however can do fine without Daz.

I was probably included in some of the last people to convert to a smartphone.
I still hate texting and email. I'll be hanging on to Daz for a bit but ultimately the writing is on the wall.
Once the bugs are worked out of AI and consistency is added, Daz will fold like a house of cards. I'm surprised Poser is still a thing. I'm sure there will be hard core "for life" fans but I'm not simping for a obsolete program

[ADAM]

@frank22

I find it odd how some people reference their Daz renders as a holy relic they attained but post nothing in the art section or fail to participate in contests.

Not directed at anyone in particular unless you think it's you then it probably is.

[Father Mapple]

the way AI currently works will never give really consistent results across multiple shots or camera angles because for this AI would need a true understanding of 3D space and memorize a shier impossible amount of details. the way it works now is more like a very skilled cartoonist or painter with an incredible ability to observe and reproduce ...

to be able to really compete with 3D it would not only need to observe and recreate, it would need to understand the observation completely like we do when we see something and from that point we are still very far away. the main limitation for that is the same as for every 3D related stuff, the limited computing power of nowadays hardware.

[Legolas18]

@stachelzelle This has actually become a thing recently, with visual understanding models.

You just feed the model one or more images, and it can do almost anything with them. It's not 100% accurate all the time (character consistency, for the best models, can reach ~90% to 95%), but if you repeat the generation a few times, it gets there eventually, unless there is something it really cannot do (like, black sclera), or is blocked from doing (like NSFW).

As for HW limitation, yes it's true, but that's why we have quantized models, to lower VRAM and computational needs at low quality cost (I am talking about Nunchaku, which is a technique that speeds up generation time by x3, at least on my PC, while lowering VRAM needs by 50%).

Example:

INPUT IMAGE:

PROMPT:
The girl is now standing and drinking orange juice

OUTPUT:

[yngvarr]

@legolas18 It's funny that your examples demonstrate the imperfection of AI. It's like a "spot the difference" game. 4 rings on the wrist of the right hand instead of three. A glove for three fingers instead of two on the left hand. A leather belt appeared on her right leg. And these problems either need to be corrected with an additional promt, or they should be ignored. So the problem of repeatability is still relevant.
The second problem is recognition. Many people who are far from 3d and AI models quickly recognize the work of AI. It's like an artist with a certain style. And, most importantly, people don't like it. I've seen a lot of comments about games with AI art like "ugh, AI..."
This leads to the third problem - And even in different AIs they are still quite similar, because AI is a great equalizer and averager.
Unfortunately, judging by the daz 5 beta, there will be no breakthrough at all. Nevertheless, daz remains the only alternative to expensive products by Reallusion and very difficult to master Blender.

[Legolas18]

@yngvarr
So the problem of repeatability is still relevant.
Only in a minor fashion and only for lazy people. What I shown was considered impossible just one year ago, and the results are already 95% there. Also, creative people use Photoshop. Which I think is good, because if AI does everything perfect, then there is no way to distinguish yourself.

The second problem is recognition
To be honest, I see people questioning real photos as AI all the time. AI moves very fast so things are changing very quickly. One the other hand, every single artistic tool also has its own distinctive "style".

And, most importantly, people don't like it.
They don't like the mass produced garbage. That's easy to produce, and requires little effort and the "average" ai user just puts so much and so quickly out on social media that it quickly loses all value. When something unique comes out it is still valued, unless someone is specifically "anti-ai" and assumes everything is AI by default.

even in different AIs they are still quite similar, because AI is a great equalizer and averager.
Totally false. There is a huge difference between different generations of AI. Some models are vastly superior to others, especially in what they can do. And, if you compare same generation models, there is a huge difference between local generation and paid services.

daz remains the only alternative...
Most people don't like most 3D renders made by Daz either.

And, if you apply what you said about AI to DAZ, or any other artistic medium, almost everything applies there too.

[trader2h]

I don't see why someone should choose between them. Both are great ways to express yourself and have fun making something you like.

I've been around since the very start of poser and when it started it was just to provide reference for painters. Now we can use DS as a reference for AI.

Most traditional artists still don't recognize 3d art as real art. Most 3d artists think DS renders are nothing but playing with dolls. And now we are in line to think AI as just cheating with no artistic value.

I happen to love AI results but also really enjoy making renders in Daz, so I just accept any medium that creates something that I like looking at.

[Legolas18]

@trader2h I agree with you, that's why now I am getting back to DAZ, to use it to guide AI art. I think the combo is really powerful at this point, especially with new AI models 🙂

[gts6]

It's like a "spot the difference" game. 4 rings on the wrist of the right hand instead of three. A glove for three fingers instead of two on the left hand. A leather belt appeared on her right leg. And these problems either need to be corrected with an additional promt, or they should be ignored. So the problem of repeatability is still relevant.

@yngvarr - Even this can be overcome. Just inpaint and and re-process certain problem areas until you get it right. Do it this section-by-section in small regions. Getting a good result without extra effort is rarely going to happen with AI. That's the main difference: with DAZ you can tweak the canvas until you get exactly what you want, and you know pretty well how it will look after rendering. With AI, you can also tweak settings, and "speculate" how it will look, but its might come close, but never exactly what you envisioned. Thats when its time to go to work to fine-tune.

[Grouchy Old Fart]

@Supreme-Leech summed it up best for me.

I still use DAZ, even today, because like Poser before it, at least half the fun is in the process. Conceptualizing, composition, tweaking, and all that goes with it is just as much fun as hitting the 'render' button.

AI could potentially be as much fun, using just prompts to get results and getting surprisingly cool output, but like others have said, it has a ways to go before it gets there.

I want character fidelity (between -all- successive images) to be 99.99% or greater, not just 95%. I say this because over a long enough comics/graphic-novel storyboard, only 95% would cause serious degradation, much like a game of 'telephone' would - just visually. Sure, if I were verbose enough or felt like re-prompting over and over again until it hits, but that's less art and more like debugging code, which is headache enough in spite of getting paid to do it.

I'd also like it to be self-contained/local, so that some online LLM doesn't take what I make and input elements of it into everyone else's stuff, causing homogenization of my style into everyone else's - and more importantly, vice-versa. Stable Diffusion allegedly does this, so it may be a solved issue?

Yeah, I'm picky. But for all its faults, DAZ (and Poser before it) had/have those two things going for it.

[ADAM]

I see it as eventually a mix that will seal the deal.

Here is a 3d scan of Scarlett.
Here is a 3d warehouse environment.
Here are all avaliable motion captures.

Tell AI: I want Scarlett to run down hallway B3 then turn right down hallway B4. While running she will look back once at the half way point and again before she turns right.

Now something that isn't animation.
Select pose picture 1
Select female picture 2
Select background picture 3
Select coloration picture 3
Select camera angle picture 1

Roll dice, poof art.

You could even have AI cameras set in a virtual environment.
Select camera 1 and place figure 2 there in pose 10a. Have them talking to figure 6 in a heated argument. Load animations for argument 4f.

It would be difficult for me to not see a world where this technology isn't utilized. Entire movies are green screen virtual environments now. Virtual people are added constantly.
Most anything can be digitally altered including lighting, faces, voices, sounds, music. Only a few things holding AI.

Why pay 50 extras when Azi can have virtual bots walking in the background.
Why pay Actors when you can buy their voice and appearance once and use them forever.

[Grouchy Old Fart]

I could totally go for something like that. If I could pull that off, I'd never have need for television again, except as occasional inspiration.

[Frank21]

Adam @d0gg0d "I find it odd how some people reference their Daz renders as a holy relic they attained but post nothing in the art section or fail to participate in contests."

Yeh but there are probably many reasons for this -
1. Shyness
2. Thinking their art isn't good enough
2. Fear of critisism
3. Too busy hoarding
4. .... add you own reason

I think everyone should at least try to post something once in a while... no pressure LOL.

[Father Mapple]

@legolas18
90 - 95% consistency (with luck) isn't enough and you will see that the last 10% will be incredibly hard to achieve, like the last 10% in quality/photorealism in 3D. that's exactly what it will make unusable for professionals and will break its neck. in the end AI will face the same problems as 3D, it needs computing power itself and if you want it to truly understand and rebuild 3D space it needs to do what current 3D programs do on top of its own computing. so it will need even more computing power than 3D programs alone.

current 3D is still a approximation and simplification because of limited computing power and this will count for AI too. there is no magic bypass for AI if it comes to accuracy and in comparison you lack the amount of control over the result you have in 3D. the 90-95% consistency wobbly morphing like aesthetics of AI may be suitable for hobbyists, but not really for professionals.

by the way, if you want to see what 90-95% consistency mean search for “Create a replica of this image. Don’t change anything” AI trend ...

“Create a replica of this image. Don’t change anything” AI trend takes off

what brings us to the next problem, the web is flooded with imperfect AI content and sooner or later this will contaminate the AI training data itself and where this leads you can see in the example above. nevertheless, the coming years will be interesting ...

[Legolas18]

@stachelzelle
I don't claim to know the future, so I cannot comment about what will be possible (or not) in the future. The only thing I know is that we went from 0% consistency to 95% in just a couple of years. Obviously, here I have just one specific tool in mind (Flux Kontext, which is mostly an image editing AI tool), but there are ways to create 100% consistent characters in a different way (with an image generation AI tool), but then I'll not have almost absolute control over that character. So, for now there are just imperfect workarounds.

As for the last 5-10% inaccuracy being a problem, I completely disagree with you. A professional having its workflow sped up by 90-95% is a huge game changer and, if they are professionals, and with all the time they saved, they can easily do the 5-10% of work needed to get the results they need in Photoshop or other program.

As for the "Create a replica of this image. Don’t change anything” that's what happens with adding noise to an image over and over. Eventually, you degrade the image.

[Author]

Posts