Design Cinema – Episode 110 – What AI Cannot Do

Author Avatar

FZDSCHOOL

Joined: May 2024
Spread the love


Design Cinema – Episode 110 – What AI Cannot Do


In this episode, Feng discusses what AI can’t do when it comes to entertainment design.

Thank you for watching. Please subscribe and help us share this video: http://bit.ly/2rF25ao

Visit FZD School: https://fzdschool.com/

Watch All Episodes of Design Cinema: http://bit.ly/2r2XFJR

source

Reviews

0 %

User Score

0 ratings
Rate This

Sharing

Leave your comment

Your email address will not be published. Required fields are marked *

48 Comments

  1. What AI can't do? It's simple — AI couldn't draw without stealing art, because for work AI need your art and arts of thousand other artists.
    Without database AI couldn't do anything. It works because internet rules allowed to stealing if it public…

  2. This doesn't change the fact that as AI gets more and more powerful, companies are willing to save 5x the costs for 15% lesser quality. The trade-off is more than worth it. It'll kill off all the lower to mid level artists, and where will they go? They'll improve themselves to compete with the top tier ones because they've no choice. So the top will get way even more competitive lol

    And that is only going to be more and more suffocating. In future versions, I totally believe AI can fix all the things Feng talks about in this video.

  3. Feng you have transcended just the Concept art world. You helped me stay grounded as a founder of a game company. Thank you for dropping another gem!

  4. Glad you're back, no matter how long your breaks, we'll always welcome you back, this episode was so cool and I totally agree on most of the points mentionned here!
    AI just can't and never will be able to do most of these things simply because it's a human thing to do, only exception is if they develop consciousness but we're nowhere near this
    As a French I kinda feel attacked there, that no one could show you good portfolio haha, I also do these big landscape small character, gonna need to do better now
    A game series I'd recommend on Switch is Xenoblade Chronicles trilogy, they kinda became the new gen "classic" JRPG and have beautiful world design and OST
    The current Persona series has that unique style thanks to Shigenori Soejima, he's just iconic and the art direction in the next game he worked on Metaphor ReFantazio is amazing, with that painterly style in the menus and character portraits, imagine if they made that with generic AI style, just no.. big no.. I hope no big game companies will ever make that mistake

  5. Hey Feng. I think your analysis of AI is missing the mark by calling it a tool like other tools and the contradiction itself is in your very analysis.

    The primary difference between AI and stuff like photoshop, photobombing etc which is the very first argument people made a couple years ago when all this hit the mainstream has been debunked not only syllogistically, but in practice as well, and that’s that unlike the other tools where you had to actually learn, acquire skill, practice, put in time, effort and understanding to produce even one image, AI has absolutely no such demands. You said it yourself, the fundamentals are absolutely not needed anymore. That is your contradiction. By definition if this “tool” doesn’t need the user to understanding anything about the process, it’s not a tool, but a slave being given general commands and it takes it from there. So you’re wrong in your assessment and you’ve not even noticed the contradiction.

    The second part where you’ve missed the mark is that unlike other tools this machine has the ability to produce artwork at an industrial rate, obliterating whole production teams in output. That kind of rate is impossible to keep up with by humans, and the cost is minuscule by comparison.

    The third part where you’re being willfully blind I think is the idea that these machines won’t be able to solve design problems. If a statistical prediction model (which is what these things are, they aren’t really AI, that’s a marketing moniker) can “learn” that ‘red’ and ‘apple’ have a correlation in color, tone, shape and put that thing in an environment with its own correlations etc etc concluding into a few billion pleasing images, then it can learn a) to create variations of the most aesthetically pleasing results based on the input of the best artists using these things currently (because as people are using these machines their aesthetic sensitivity is being taken advantage of to train these machines even more on that skill) b) it can learn to be independent of the input of a user based on the same principle of learning from the input of those using it and c) it will be able to learn to solve design problems based on the previous two principles.

    So, yes your students have design skills, yes the are getting more work, yes they are highly paid. But since they won’t need to know fundamentals any more, won’t need to know how to draw and paint all that well if at all, won’t need to study that much, the number of them will increase because the level of entry will be lowered, the ease of learning will be significantly smaller, and since the supply will become larger the jobs and money will decrease as well.

    But even then that’s a optimistic scenario because these machines learn at a geometric rate Feng. The explosion of things they can do based on huge swaths of data which was the key shows that if you have the data and the processing power it’s only a matter of time, short amount of time, until 90% of the problem, be it drawing, painting, composition, design is solved. The first three more or less have been solved, the last one is what’s left.

    Given that it seems the time your students will enjoy in the sun is very short, because the key phrase here is they are in demand and well paid because of AI… for now.

  6. What's crazy is I randomly decided to look up this channel just now after learning art and design over 5 years ago on this channel. Amazing content again!

    100% agree that AI can never solve design that conforms to the real world interactions because things need to fit perfectly otherwise it falls apart

  7. 1:16:39 To spring off of this discussion, Panty & Stocking, a Japanese animated show that inspired a generation of artists (milennials and old zoomers), animators and art direction was created because a bunch of drunk artists at Studio Gainax made a bet that they could make a more interesting Western cartoon than Drawn Together. Whether they did that is objective, but it's one of my favorite production stories ever.

    Humans are capable of making dumb ideas interesting and inspiring from just socializing, bouncing ideas off each other and editing them later. That’s a soft skill that no amount of sitting alone behind a computer and repeatedly prompting can teach.

  8. I think the main problem with Ai is when companies like Midjourney and Stable Diffusion train their Ai by stealing work without compensation or consent to add to their data set. Then because its free and open source people take that and some purposely copy/mimic artists and sell it for profit. The ethics behind is what's wrong, copyright, etc. The tech itself is cool but the companies need to go back to the drawing board.

  9. So glad these videos are back! I'm super interested in learning 3D/Blender for the purpose of 2D art and design. If there are any existing FZD resources on that, I'd love to start learning.

  10. Theres 2 things Ai will never do. First person conscious soul aware experience and judgement. Funny thing is, any entertainment is a byproduct of life. So it has to happen to be told. You dont get to watch Saving private Ryan without it happening, it has to happen first. Life is a test, you have to go throught it. God bless.

  11. At 40:00 Feng mentions that at a senior level concept artists get to do the fun, flashy images. I'm wondering does that make senior people less employable in the future if their portfolios are filled with key art and box covers? Say if they are made redundant?

  12. I guess the question is will artificial intelligence have the proper building blocks to achieve agi? Are these algorithms and our own consciousness or sentience really on the same plane such that an actual agi being will be able to do this too. The statements made in this video are true for gen ai and diffusion but we may be wrong if agi is achieved. Assuming this isn't some dead end pursuit to bring in investor money.

  13. When the stigma of using AI is past… more jobs will be lost. Right now, companies are not "all in" on AI as the tools are still improving and the stigma is still risky to deal with. Both of those issues will be gone soon. Also, the current business models are built on a pre-AI budget. It will not take long for those budgets and the stigma to adjust to the reality of AI. When that happens, watch out. That is when the industry will really change and artists in this field will have to make some significant changes.

    AI will be incomprehensibly better in a couple years time. When that is the case, everyone will be using it. The artist will have to shift from being a "drafts-person/designer" to something more akin to a director. It will create a new type of artist.

  14. People should view AI as a timeline of eventualities and adjust their careers accordingly. Eventually it will be able to do all the things listed here on the "can't do" list. Not only that but it exceed expectations because of emergent properties. It will be programmed to execute functional looking components, it will design multiple elements properly within a scene and it will do these things faster and better.

  15. Thanks for sharing your immense knowledge on the deep design process!
    Intrinsically the technical, industrial design understanding to incorporate more realism via function is crucial.
    There are some new advancements with AI like ipadapter/LORAS than can be specifically trained on those cars or characters you mentioned to be integrated into existing designs.
    I'm sure there'll eventually be fine-tune models and workflows catered to the inside world of an IP.

  16. I really appreciate that you didn't trash AI generated images and that you gave them at least some credit. On the other hand I believe that up to 80 percent problems that you've mentioned could be solved in next ten years especially with interconection between LLMs and image genereation.

  17. as good of a breakdown it is , i cant help but notice the feeling selling us career as a designer, functional design is very different from visual art , u don't even need to know a single drawing or painting fundamental to think about functional design more less with upcoming genAi surge.

  18. Just to add a little, what I've heard recently is illustrators who do commercial work for agencies are seeing their work decimated by ai. I've seen pitch decks full of ai work on every job since January. Shooting boards are still fine because directors need the control and a quick turnaround, but more rendered illustrations for pitches are almost dead. In the past they would sometimes use stock photos for the pitch decks, but I'm not seeing that much either. For the record, these ai images look extremely generic and they all have the same color pallet, but the agencies don't care because it still gets the from A to Z.

  19. For the refinement part, you could take the generated image and paint on top of it. Then use a mask on the area that you painted over and a low denoise value to regenerate part of the image. Though, the result may or may not match the overall style of the entire image.