Is AI killing the 3D star?

1567911

Comments

  • vonHobovonHobo Posts: 1,681
    edited November 2023

    Federmann said:

    Is AI killing the 3D star?

    Only very recently I have started to ‘publish’ some of my work on DeviantArt. Doing that I also made my personal assessment of the state of affairs on that platform. It would seem that AI is crushing it. Despite the lack of narrative continuity AI imagery seems way quicker to generate ‘likes’/’favs’ than carefully conceptualized and composed 3d renders. Some 3d artists that I liked said they were feeling the crunch in their already modest online revenue streams since the advent of AI (‘nobody is interested in story lines anymore’). As for AI, I keep hearing that 2024 will have AI that is so much better (and I am already blown away with what is possible now). All this won’t keep me from plodding away at 3d especially since I want to do some (minimalistic) animation to accompany my music but I now realize that new kid in town will soon own the town except for a few 3d hovels near the (burning) forest.

    I hope you disagree!

    PS: I have already discovered it is not necessarily an either/or question; I have created some very nice (AI) BGs to be used in 3D renders and AI 'can come up'  with great ideas for architecture, props, clothes etc.

    I'm a traditional artist and sell prints of my work online. Business has never been this bad.

    I asked someone if they thought it could be AI generated art that is causing sales to go down for traditional, CGI, and 3D artists. And they said " anyone could always copy and print any art they wanted online". So there is that argument.

    Could it be that inflation is so high that most people can't afford luxuries like an art purchase lately??

    So I am thinking it's a combination of many factors... inflation, no more govt assistance covid checks, no more need for home offices and the wall decor to go with it, no more being stuck inside and restricted to shopping online, more people going on expensive vacations again finally rather than spending that money on other things, and the ease of anyone being able to create art with AI now of course is always on my mind as the AI gets better and better at that sort of thing.

    We have to beat the machines by turning their tactics against them.

    Do you remember the episode called "Nomad" from the original Star Trek television series?

    The robot called Nomad boarded the Starship Enterprise, it's mission to destroy all imperfection. That meant destroying humans which it deemed imperfect.

    James Kirk finally outwitted the robot by posing a simple rhetorical question. NOMAD overheated when it went into an endless loop trying to answer the question, and then it short-circuited, and blew up!

    We can beat the machines at their own game!

    Nomad (voiced by Vic Perrin), Mr. Spock (Leonard Nimoy), and Capt. Kirk ...

    Post edited by vonHobo on
  • fred9803fred9803 Posts: 1,564

    I totally understand you AllenArt. I was strongly opposed to AI (and said so) when it first appeared, mainly due to the "stealing" thing. But in reality the art styles are mostly 100+ years old and therefor public domain. Things like the Pixar and Disney styles etc. are perhaps dodgy, but the final rendered images probably have anough human input and manipulation to skirt IP claims. Likenesses of celebrities is another thing, but if the raw images were taken in a public place I guess they'd be OK, unless you do something naughty with them, and publish the image, a seperate issue.

    I haven't used DS for almost a year now and to be honest I do miss it and the process of using 3D. Been doing it for 20+ years. It's just that I've always been fascinated with achieving realism and DS just does't cut in anymore in compaison to SD. With regard to hair, clothing, faces and lighting, AI is just so much better at that. 

  • This "stealing art was ok because we couldn't afford to pay for it, and it was just for research ( aka commercial enterprise )" really annoys me. If that was ok, would it be ok for me to steal a car so I can learn to drive, because i can't afford to buy one? Or for me to buy and use  a car I know was stolen because "how could I possibly compensate the original owner"?  ( ok with cars there are ways to do that, but the principle is the same ).

  • WonderlandWonderland Posts: 6,873
    edited November 2023

    I'm having fun taking renders into AI phone apps! 
     

    First one is a straight render of Mira 9 (with a lot of tweaking done on the teeth to give her a decent smile!) and other new products I was going to post in the render thread but now I forgot their names. Anyway, the others are AI of that render and look MUCH better than the original. AI is always going to look better but with 3D we can at least create the main character and poses but now that I see how good the AI hair looks, I may stop buying hair! The AI created clothes with no input from me, just came as presets though, no way to control it as far as I can see in that app (Facetune.)

    'BUT, many still have big flaws, and even the last one you can see the wrist is funky and the bangs on the hair are weird. Also clothes cover up all cleavage on that app. No way to do nudity or sexiness that I have found with AI. But it's great for portraits!


     

    g9 mira 80 protostar hair pose 3 full scene dofcbnd dress pose 4cam 2.png
    1200 x 1600 - 3M
    Facetune_27-11-2023-11-28-18.jpeg
    1200 x 1600 - 385K
    Facetune_27-11-2023-11-40-53.jpeg
    1200 x 1600 - 341K
    Facetune_27-11-2023-11-32-41.jpeg
    1200 x 1600 - 384K
    Post edited by Wonderland on
  • ArtiniArtini Posts: 9,456
    edited November 2023

    Valid point - AI is not for everyone and we need to accept such preferences,

    but for someone else, it could be a different story.

     

    Post edited by Artini on
  • frank0314frank0314 Posts: 14,049
    edited November 2023

    Artini said:

    This is the answer, about how to connect Daz Studio with AI:

    https://www.daz3d.com/face-transfer-2

    I don't believe this is true. The AI for DS that Daz has discussed has to do with setting up scenes.

    Post edited by frank0314 on
  • ArtiniArtini Posts: 9,456
    edited November 2023

    Sorry, I am exagerrated it a little, but it is something, that confirm some interests of Daz 3D,

    and I like their connection to the new technology very much.

     

    Post edited by Artini on
  • generalgameplayinggeneralgameplaying Posts: 517
    edited November 2023

    background said:

    This "stealing art was ok because we couldn't afford to pay for it, and it was just for research ( aka commercial enterprise )" really annoys me. If that was ok, would it be ok for me to steal a car so I can learn to drive, because i can't afford to buy one? Or for me to buy and use  a car I know was stolen because "how could I possibly compensate the original owner"?  ( ok with cars there are ways to do that, but the principle is the same ).

    Fair use on everything for commercial purposes is a wet dream of some people. Even wetter with ai staying in the cloud, and the laptop never seeing the training.

    The rain will flood the grounds beneath anyway.

    The ai systems of the time clearly are of encyclopedic nature, and it would be a big mistake to let them have fair use for training. Also the TOS trickery of some platforms, should fall into the same category.

    However, eventually, some will have curated data sets, and maybe some cc-type-of licenses with ai training allowance will have spread to an extent, that you could train a system.

     

    Just a moment ago, some players just tried to demonstate the capabilities, as cheap as possible, and some players tried to profit as fast as possible. For smaller players, the cloud will become shifty terrain pretty quickly. Either big players will dwarf them anyway, or the whole technique will go open source and "laptop" in sufficient quality. They are short-lived on that account, though at present, most are funded by venture capital, if i read this right.

    Post edited by generalgameplaying on
  • WendyLuvsCatzWendyLuvsCatz Posts: 38,208
    edited December 2023

    https://developer.nvidia.com/blog/generative-ai-research-spotlight-demystifying-diffusion-based-models/

    interesting explaination of diffusion models

    no I  don't understand it either cheeky

    Post edited by WendyLuvsCatz on
  • kyoto kidkyoto kid Posts: 41,044

    ...I have my gallery blocked to AI. and disabled downloads.

  • kyoto kid said:

    ...I have my gallery blocked to AI. and disabled downloads.

    Yes, same. Initially the users had to go to each posted piccture and apply noai in a short timeframe. With over 600 pictures posted back then and a full time job irl this was not possible.

  • artxtapartxtap Posts: 158
    edited December 2023

    Jabba said:

    I have combined 3D, AI, and painting in what I suppose we'd call mixed media images... and yes, I'm totally blown away with the results (mainly days/weeks of time saved introducing AI for complex backgrounds, using 3D for precision items and then painting the whole thing pretty much like I would postwork many purely 3D renders).

    The problem as I see it is that non-artists see how impressive AI art can be and there appears to be a trend of some folks feeling that "art takes no real effort or proper skill, and therefore has no proper value", and that will mean lean times for artists trying to use their work to pay bills.

    Of course, if you only do art as a hobby, then none of this makes one blind bit of difference to you unless you want to experiment with the "evil AI" yourself, hehehe.

    …...back in the day….way, WAY , WAY back in the day programs for devices, computers, palms, blackberries were, expen$$$ive, but TOOLS came along that made the process much easier. Than come the ‘smart phone’ and android.. programs prices DROPPED !!! I ported my programs from Palm to android because the TOOLS were free, eclipse, android, etc. now anyone can code(cook) because the TOOLS are there. AI is a TOOL and yes like you said this TOOL will effect the price of ~art~, just like other TOOLS effected the price of ~computer~ programs. 

    Post edited by artxtap on
  • ArtAngelArtAngel Posts: 1,692

    My comment may be perceived as tunnel vision, but to those of you who don't know me, which is probably everyone here and everywhere else on earth (reclusive me) I think this: An editor can enhance a writer but never replace one. I see AI as a bad editor where you need to hire another editor to edit the AI work. I see AI as bringing a really cool perspective on inspiration. Just another source for inspiration to those of us who need it. I have never used AI but find when

    Artini uses it

    it does produce some jaw dropping visuals.
  • kyoto kidkyoto kid Posts: 41,044

    ...I see the difference between AI generated art and physically creating a work (even using Daz) sort of like the difference between digital photography and film photography. 

    With digital cameras you can just keep shooting until you get the results you want and simply delete all the rest.  With film getting that perfect photo is often much more of a challenge and requires a keen eye as once the shutter is snapped that's it.  Yes there were motor drives and film magazines, but that was expensive for the hobbyist photographer, not just for the equipment but also film stock and processing.  This made getting that one perfect photograph where the lighting, framing, subject, and everything else just "clicks" together, gives one an incredible feeling of satisfaction and accomplishment.

    That's why I have no interest in AI generated  art. as all the "heavy lifting" is done by algorithms not one's own mind, skill, and imagination.

    This is why I liken it to the old Daz "Make Art Button". as one doesn't need any artistic background, skill, or training to produce a viable image.

     

  • grinch2901grinch2901 Posts: 1,246

    I don't know what art is, really, It's whatever comes from whatever creative process you used to produce it, and even then there will be others who say 'it's not actual art'.  Heck, look around Reddit and you will find plenty of people that say Bob Ross didn't make art, and that's genuine paint applied to canvas so the media isn't enough to say it is/isn't art. 

    Here's what I do know: my stuff isn't high art, it's just for fun and personal use. For 17 years or so, my tools of choice were Hexagon and Daz, but I was getting burned out on DAZ and using it less after G9 came along. Step in the AI tools to fill the gap, midjourney and stable diffusion, just goofing around.  But whereas in DAZ I had total control of the scene I wanted to use, in AI I have infinite possibilities but nor a lot of control, whcih is frustrating.   

    So lately I've been coming back to DAZ as part of a pipeline where I set up my character in DAZ with DAZ content and then pass the render to Stable Diffusion to let me have a character I want (combo of DAZ and Stable Diffusuin), dressed and posed exactly as I want (all done using DAZ / DAZ content), rendered out in a setting  and style that I want (Stable Diffusion, but the backgrounds are still not that good in my current pipeline).  For me it's just a fun distraction anyway, I don't make money off it and rarely even show it to anyone.  Honestly, I think that my recent fliration with AI tools has been making me MORE inclined to come back and use DAZ again, this time the middle part of a process instead of the end of it.

  • ArtiniArtini Posts: 9,456
    edited December 2023

    I am also playing more with Daz Studio recently, especially after the release of Face Transfer 2,

    which provide me a lot of fun playing around with AI generated portraits

    and then transferring them to faces of Genesis 9.

    I do not regret playing with Stable Diffusion and the other AI tools

    because they provide me with the training/interest to my aging brain.

     

    Post edited by Artini on
  • The lack of control of AI continues to prevent me from anything but incidental usage of those products. I've been able to usefully incorporate some AI-generated images as wall and table art within a Daz scene, but a few days ago, I just needed a quick outdoor street scene to be seen through a distant open window. I considered using one of the Daz street assets that I have, but it's such a small part of the image that it wasn't worth the time of messing with it, so off I went thinking I could whip up something in AI in a couple of minutes. After 15-20 minutes of not being able to get a single correctly oriented street, I gave up in disgust and put up curtains. I think AI knows that I hate it.

  • mike.smith2005mike.smith2005 Posts: 85
    edited December 2023

    Like with anything else, AI can be a tool you can use as part of your workflow. "Real painters would never use spray paint." "Real painters would never use a projector." "Real painters would never use 3D modelling." Anything new first comes on as a crutch as people overuse it and rely on it in place of skill and talent and training, Eventually, the fad subsides and people begin to see how it can be used to enhance certain aspects to production. AI might make sense to de-age a portrait of a person as part of a larger work. AI might make sense to create a texture for an alien landscape. AI might make sense when converting items of scale in composition.
    But typing a prompt into a field and hitting "generate" over and over until you get something you like isn't creating art. At best, you are producing content. It can be part of the process but not the final result.

    Post edited by mike.smith2005 on
  • NylonGirlNylonGirl Posts: 1,809

    HLEET_3D said:

    It will kill all the small artists business on zbrush for fanarts XD along the way. 

    Kiling the small artists would kill the AI because the AI won't have any source to draw from. 

  • WonderlandWonderland Posts: 6,873

    I think if you bring your own renders into AI (or your own photos) it is perfectly fine and not "stealing" anything from anybody. Although I still buy Daz sets and props which is great for having the character interact with the set and objects in the scene, I can easily make quick AI backgrounds if needed. People were buying pre-made backgrounds snyway for their art, mostly on Rendo, but I think there are a few here. You could probably even stick some onto existing products that come with a backdrop like some of Dreamlight or Colm Jackson's products. 
     

    I honestly don't think AI is stealing from anyone unless you are a truly famous artist like Van Gogh where it can put his style on your art. I have lots of art online with my name in the title so it can be found on google and when I tried to get art in my style or mentioned my name in the prompt, I got art that was nowhere near what I do but did see scrambled watermarks that looked similar to my font. It read the text of my name on my watermark and stuck a scrambled version of it onto an unrelated image.  So unless you are a famous artist, it won't even pick up on your style. Try as an experiment putting your name in the prompt and see what happens. Anyway, I think it's totally fine to put your renders through selfie apps. I've gotten some great results. And Photoshop has even incorporated AI, legitimizing it. Tafi itself is creating an AI version for creating avatars so even Daz is OK with it. 

  • WonderlandWonderland Posts: 6,873
    edited December 2023

    Two more examples of how AI enhanced my renders. The faces are the same but enhanced the way a human selfie might be. It wasn't all automatic, I had controls. You can even change the makeup and lighting. It's basically post work. The original hair product in the second one I don't like and was planning on returning but I was able to completely transform it with a selfie app by actually putting in a prompt of how I wanted the hair changed. It's not stealing anyone's hair, it just, often drastically, changes the original hair. I think I put in "long layered auburn' and tada! I didn't create the original hair anyway, I purchased it here. . I think AI is brilliant and will be using it on most of my renders from now on! 
     


     


     

     

    Post edited by Wonderland on
  • NylonGirlNylonGirl Posts: 1,809

    Wonderland said:

    Two more examples of how AI enhanced my renders. The faces are the same but enhanced the way a human selfie might be. It wasn't all automatic, I had controls. You can even change the makeup and lighting. It's basically post work. The original hair product in the second one I don't like and was planning on returning but I was able to completely transform it with a selfie app by actually putting in a prompt of how I wanted the hair changed. It's not stealing anyone's hair, it just, often drastically, changes the original hair. I think I put in "long layered auburn' and tada! I didn't create the original hair anyway, I purchased it here. . I think AI is brilliant and will be using it on most of my renders from now on! 
     


     


     

     

    This is kind of what I'm taling about though. I was saying the AI can't replace small artists because it needs them to draw from. In this case, you are the small artist feeding the AI your own work. My comment isn't meant to say the AI is bad. Just it can't eliminate the artists because it needs them, even if the artist is a willing participant. 

  • grinch2901grinch2901 Posts: 1,246

    NylonGirl said:

    HLEET_3D said:

    It will kill all the small artists business on zbrush for fanarts XD along the way. 

    Kiling the small artists would kill the AI because the AI won't have any source to draw from. 

    A trend in the AI world is the use of synthetic data to train the models. That is, rather then using primary sources, you use the best models available to generate data and then use that to train other models. Apparantly according the academic papers, it works well and from a business perspective it abstracts that direct link to copyrighted works. So win-win I guess.  What would this mean for AI art?  I'm not sure but I suppose all the stuff we, the public, generate in a program lid Midjourney would form the database for training the next models, not real works from real artists.  It will be interesting to see what that does to the images Midjourney creates, will it make them more flat and dull or somehow more wild and surreal and imaginative?  I imagine time will tell.

  • FSMCDesignsFSMCDesigns Posts: 12,754

    NylonGirl said:

    HLEET_3D said:

    It will kill all the small artists business on zbrush for fanarts XD along the way. 

    Kiling the small artists would kill the AI because the AI won't have any source to draw from. 

    Agreed. If every AI tool was denied access to any image on the internet and could never have access to any images on the internet, I would totally be on board the AI bandwagon as a tool. As proven time and timae again you can't rely on users to do the right thing, so eliminating access is the only way.. But then again, many of the amazing AI images users want to emulate would never exist..

  • JoeQuickJoeQuick Posts: 1,704
    edited December 2023

    I appologize if I've posted these before.  I really enjoyed stable diffusion for a stretch over the summer, where I used it to re render my old Great Greymalkin robot in a variety of styles, using in-painting to refine certain regions until I was happy with them.  That was like my second or third "phase" of AI.  My first couple phases were using it to gussy up old promos or to concept creatures which I'd go on to model. 

    Though, the concepting thing was always a bit of a gut punch, because I'd know that whatever I wound up with was never quite as good as what Midjourney spit out. 

    At another point, I got into using stable diffusion to convert niji midourney renders to pseudo 3d ones.

    But, like anything, each of these things feels like a phase.  I do it for a time, but eventually it gets tired.

    00012-1694360081.png
    1380 x 1800 - 3M
    00050-2060175781finaler.png
    1392 x 1800 - 3M
    00004-4191365999.png
    1368 x 1800 - 3M
    00011-1566430100.png
    1380 x 1800 - 3M
    00023-3469181615final.png
    1380 x 1800 - 3M
    00062-2060175781.png
    696 x 896 - 875K
    00034-2172402054final2b.png
    2048 x 1144 - 3M
    Post edited by JoeQuick on
  • kyoto kidkyoto kid Posts: 41,044
    edited December 2023

    ...I agree with mike.smith2005. as a tool in your workflow it's one thing.  Using it in post is doing just that. it's just another filter. What those like myself have issues with for is the generative version that uses the work of others for the learning process without compensation and has the algorithm do all the "heavy lifting.  Any couch potato can type in a string of random and press [enter].  For what we do still takes leaning, on our part skill, and effort to produce a scene from scratch, more so if we also do our one modelling kitbashing sculpting, or texturing,

    I still remember the days when we had to write code to create a simple 3D graphic image. 

    Post edited by kyoto kid on
  • gooHerogooHero Posts: 16
    edited December 2023

    In some way AI art is definitely gaining ground and getting better fast; invalidating some artists work entirely, because they have an almost identical art style.

    For me personally, it doesn't have that large an effect though, as my themes are very specific and very few AI model trainers go for them. The problem with AI for artists that live off of their work is real though, I think.

    On the flipside AI "art" is not enticing or interesting to me in any way and to me as it all looks the same (in the erotic genre, I don't look at the rest or "art" in general as it has no value for me). I can, as probably most of you can too, identify AI "art" inside seconds or even milliseconds of clicking a thumbnail - or even from the thumbnail itself.

    In a few years or maybe even next year, another dilemma will appear, because then all the low effort "artists" releasing their "work" will also lose their shine in the limelight, as every joe will be able to create content for themselves inside a day and aren't reliant on them anymore, because of the ease of use. I trained a model in 6-12 hours I think it was, and I was creating images inside a day. They sucked, but they were there. In the end it will fizzle out and noone will give a [darn] about others' AI "art" as you can just make what you want yourself with minimal investment ($10-20) and time.

    regards,

    Anti

    (Edited by mod for language)

    Post edited by frank0314 on
  • Training ai with generated content. Well, so far that's more like training a detection ai with content produced by a generative ai. I am not aware of easy ways to train a generative ai with generated data - there are some, if you start from 3d-models rendered from different angles, but doing it for "everything" pretty much resembles DAZ repairing all meta data and descriptions by hand, or for an ai training data set, having it tagged correctly from start (by magic). I could imagine combining tagged images from the wild with "generated" content that helps defining some thinigs better statistically. But that doesn't appear to be generative ai as input to generative ai, at least not in the general form. I could distantly imagine, that you could train an ai, capable of doing some specific things very well, but not the whole thing, and employ it to more or less dynamically create more training data in specific contexts (where it has to fit, which someone or something has to detect or know beforehand), and thus improve statistical odds for useful output in those specialist areas. Be it drawing hands, humans, horizons...

    Oh and careful with the short cut (~ short circuit): https://www.cdotrends.com/story/18288/training-ai-outputs-generative-ai-mad

    The competition to it's input is an issue, after all.

  • grinch2901grinch2901 Posts: 1,246

    generalgameplaying said:

    Training ai with generated content. Well, so far that's more like training a detection ai with content produced by a generative ai. I am not aware of easy ways to train a generative ai with generated data - there are some, if you start from 3d-models rendered from different angles, but doing it for "everything" pretty much resembles DAZ repairing all meta data and descriptions by hand, or for an ai training data set, having it tagged correctly from start (by magic). I could imagine combining tagged images from the wild with "generated" content that helps defining some thinigs better statistically. But that doesn't appear to be generative ai as input to generative ai, at least not in the general form. I could distantly imagine, that you could train an ai, capable of doing some specific things very well, but not the whole thing, and employ it to more or less dynamically create more training data in specific contexts (where it has to fit, which someone or something has to detect or know beforehand), and thus improve statistical odds for useful output in those specialist areas. Be it drawing hands, humans, horizons...

    Oh and careful with the short cut (~ short circuit): https://www.cdotrends.com/story/18288/training-ai-outputs-generative-ai-mad

    The competition to it's input is an issue, after all.

     

    In the word of AI text generation models (Large Langage Models, LLMs) Microsoft publisned their "Orca" paper and model that used Chat GPT to guide the training, so as part of the training the model was interacting with ChatGPT to see if it got a good answer and then learning what worked/didn't work. 

    https://www.microsoft.com/en-us/research/publication/orca-progressive-learning-from-complex-explanation-traces-of-gpt-4/

    Could this approach be used for imagery?  Maybe, if something like ChatGPT was trained to recognize styles, for example, and be able to say 'not quite there yet' or 'oh yeah, spot on!'. Right now the synthetic data thing is more about the textgen stuff, not images.  However I think the big players want to get in front of this lawsuit stuff so if they can train the models using expert guidance from an AI to create synthetic data instead of ingesting copyrighted data I think they will be motivated to do it becore the hammer of the law comes down on them. 

  • ArtiniArtini Posts: 9,456
    edited December 2023

    I am waiting for the next step in generative AI, when no reference images are necessary at all.

    Such concepts are already coming to self driving AI vehicles,

    where AI is just learning by itself with no acess to human behaviors in traffic.

     

     

    Post edited by Artini on
Sign In or Register to comment.