when the world needed him most, he returned..

Design Cinema 110: Return of The King

The mere amount of comments of people so genuinely happy to see you back is also something AI will never have.

I remember Feng making a video like 8 years ago where he almost word for word described Midjourney and similar AI art programs in a theoretical way, and everyone in the comments was saying "this is insane, wtf are you talking about, that's going to take like 400 years" and now here we are

I used to watch your videos religiously when I was a teen preparing to enter a design school. Now 10 years later I'm able to support myself as a freelancer and I look back to you as one of my biggest influences in terms of design-thinking.

My eyes must be playing some kind of trick, last time a design Cinema dropped i couldnt even draw a box in perspective. I will remember this day!

The most therapeutic video for concept artists anxious about AI

One of the most sober down-to-earth analysis of AI I've seen. No doom and gloom, no pessimism, just a realistic observation. Skilled artists who know how to design and storytell aren't being replaced anytime soon.

FZD IS BACK!!! Wooo we love this series

it feels surreal to suddenly get an upload after 2 years of silence

Literally been ages since he posted, Feng was such a big part of me developing as an artist but also a person in my early twenties. Glad to see him back!

Glad you're back Feng! Probably the most sober take on AI so far

Holy it's been years...

I instantly let go of everything else and began absorbing the new episode

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!

Finally, he uploaded again, especially after rewatching another 200 episodes for like 100 times already.

Sir! It's an honor to have you back!

Feng I know your channel is about concept art, but I think a lot of illustrators watch your content too (me included). Would you be able to provide a perspective from the illustration side of things? You're one of the few KZfaq channels that go this in depth for the entertainment art industry & I think a lot of us would love to hear your opinions on that too.

Thank you so much for caring about us Feng

I am watching Design Cinema since the beginning and I am in the industry mostly because of the amount of knowledge I gathered from this. Thank you Feng Zhu!

Hi Feng, Thanks for putting this out. I think Jake Parker and yourself are currently really helping the art industry to realize that WE have the power to change the conversation on what AI can and cannot do. Excellent break down. I have figured out a few from working with my clients, the main one for me that I voiced on LinkedIn last year was, doing art for fun and no amount of typing will replace that for me. I'm really glad you nailed it with this detailed list. Glad to have you back.

Yessssss FZD IS BACK

I rubbed my eyes when I got the notification. I couldn't have believed it if I had not seen it.

Im no longer interested in doing concept art. But always great to see Fei upload. These videos were my favorite thing back in high school, and a main source of inspiration.

what the crap!?!?!?!?! i missed you so much feng!!!

Really great and interesting video. I haven't watched Design Cinema for a while, but these videos definitely helped inspire me. On the topic of these "AI" models improving one day, I feel that they won't really get that much better, in fact they may get worse. All of these models essentially interpolate/mix distinct data points to produce their outputs. As an exaggerated example, they take 10% of image A, 10% B, 20% C, and so on mixing these similar data points based on the prompt. The scale there is exaggerated, as they're actually mixing tens of millions of data points, so maybe it's actually something like 0.00000001% image A, 000000000.2 % B. It's this extremely fine scale that fools people into thinking a brand-new image has been generated, people probably wouldn't be fooled if the models were smaller. I guess five billion images is too fine a mix to imagine. The fundamental why in which these diffusion models work, essentially guarantees that they can't get much better, companies creating these models have already scraped pretty much the entire internet, they've run out of data points to train on. These models are pretty much unable to generalise outside their training data, in fact one of primary purposes of diffusion models is to accurately reproduce their training data, essentially functioning as a highly advanced compression algorithm that is able to compress all the image on their internet into a few gigs. Since the models cannot generalise, and they've pretty much run out of real data to train on, the only method to improve would be algorithmic, but diffusion models like Midjourney, Stable Diffusion, and the new Sora (probably trained on the entirety of KZfaq btw), but as we've seen over the past year or so, outputs from these models have generally plateaued. What I'm trying to say here is that, if significant improvements to the algorithm were easy and coming as they have been saying for a while, e.g. "it'll be able to do interiors", "it'll figure out perspective", "it'll understand anatomy", then I think we might have seen something more by now. The fact is, it will never understand perspective, anatomy, function, or anything that current human artists do, even at a basic level, because the fundamental function of these models is to accurately reproduce training data, making new outputs from that is just a quirk of that function. You might even call it a hallucination, or glitch. Also, quickly on the topic of training on their own synthetically produce data to make infinite training data, firstly: if the current naturally trained models cannot generalise to every possible needed output, then they cannot generate new data to fill that gap. Secondly: these models are trained by picking up patterns in pixels of images that they associate with a text input, the patterns are so imperceptibly tiny we cannot register them as human, but these models can; so here's the question, when a model that reproduces and outputs patterns in a notable pattern detects the same generated pattern in a significant pool of it's training data, what will happen? Well it's not good, I think eventually it would just begin to output some weird Gaussian looking noise no matter what the prompt. I think the companies recognise this problem, so they are now starting to implement tagging for artificially generated content on the internet, probably to let their scraper bots screen out this data during the next training run. Lastly, while I recognise that AI in some forms could be a handy tool, this AI is not meant to be. Unlike the advents of the camera, Photoshop, digital sculpting and painting, accessible 3D, and photobashing, this tool is meant as a direct and complete replacement, the fact that it actually can't, doesn't change the intent of the companies or that of the people intending to use it. These models were trained on artists works without consent in an attempt to become middlemen in entertainment production and cut out the that people are essentially responsible for how good the models currently are. That's just the start, many people involved in this scene highly recommend prompting in specific artist names to mimic their style, creating images that look quite similar to said artist without involving them at all. This cuts off revenue, screws up SEO, and can be used for things that the artist never would have agreed to, it pretty much usurps their artistic identity. If you need a specific example, check Greg Rutkowski, he's probably the model prompted artist. These models are incapable, built on misinformation and unethical practices. I don't recommend anyone use them in their current state. On a lighter note, sorry for the very long comment, it's just that this is a big topic and I think I've made my current stance on AI clear. I hope anyone reading found it interesting at least, if you're reading this, have a good day :)

Its cool to see your perspective, Although I would argue for caution here when saying AI can't be trained for this, spatially aware generative will come out in not too long and they do open a path for AI to be trained in functional design through self corrective learning, these models seem to learn very well how physical models interact and how they need to change in order to fit together etc. This is essentially the main problem being solved by companies building humanoid robots and they solve the infant square peg round hole problem very fast. The next problem up is solving a Rubik's cube without being specifically programmed for it but rather trained through physical experience. They will likely solve that one as well, and lets be honest if you can get a physical understanding of the mechanics behind a Rubik's cube you can likely solve any technical design problem needed for entertainment. Next level up from there is mechanical engineering, and so long as its possible to train it well in understanding physics and manufacturing in the same way as you can train it to do other physical tasks I wouldn't look to far before id say it can do that also. Sure this isn't happening right now, but it will, and probably sooner than we would like.

Thanks god you’re here!!! 3 years I every month went to this channel in the dream of seeing here a new episode! AND IT HAPPENS!!!🎉🎉🎉

Return of the king! I started watching you in university, now I'm well into my career.

A couple days ago, after many years, I just decided to watch some Design Cinema episodes. And today, A NEW ONE DROPS! This is awesome.

I've been checking the channel frequently for the past two years hoping for a comeback. Glad to hear you had a good year's rest and so glad to see you upload again, looking forward to episode 111!

Much welcome episode of Design Cinema! Thank you for all your hard work!

Wow, you are back!!! ❤ we love you Feng, and we missed you

Time is up! Class Just started! 🗣🗣🗣

So happy that you're back! The series has returned :) thank you so much for putting this together.

I started watching FZD a few months ago and loved the videos and was just happy that they were recorded and left on YT. I NEVER thought that Feng would make new videos. Thank you so much for everything you've done!

Yo Feng man, It is SO GOOD to see you back! When I last watched you I was aspiring to get into concept art. And now I am actually in a course finally learning the ropes. So much of your advice has worked and I will still keep using it in the future. I want to thank you again on behalf of everyone here how much you have been able to give us FOR FREE and accelerating all oure progress.

holy shit YEARS HAVE PASSED im so happy to see FZD back on my sub box :D

I MISSED THIS FENG! Thank you for returning!

So happy to have you back! Thanks a lot!

Feels amazing to have you back! You share so much knowledge and sparks my brain. Ive have been from the very first episode and you have been the best guidance so far in my career. Working now at a top company in mobile game development. Thank you from the bottom of my heart

Glad you are finally back!

Welcome back! So happy to see more from Design Cinema, your videos are always such a wealth of inspiration and information!

The master returns! Honestly the best video I've watched on KZfaq in years and long may it continue. Cheers Feng! I cant tell you how appreciated Design Cinema is to so many of us. Welcome back!

My own research on this topic: AI is also bad at emulating the gesture of paint strokes, and the energy in an inked or penciled line. Same for overlaid brush strokes with opacity. All is always "melted" and color mix and do perfect local gradients or hard edges. So I started to increase that part of my work and show more work in progress on social media. But thanks to your video, I realize that I was focusing too much on the rendering part of AI, and I overlooked the functional design part, and only had a vague intuition that this tool was not capable of composing and resolving complex scenes with storytelling elements. So, thank you very much for your video and all I have learned about these aspects.

Finally remembered your youtube password I see. On a more serious note, glad you're finally posting again!😁

Thanks for return Feng! A lot of fresh air with this episode, I was hoping this for so long

I just can't believe this is back. This makes me so happy. Thank you ever so much!

Sad thing is that a lot of concept art doesn't need to be that functional. I've been doing mostly character, creature and skin concept design so far and most of those jobs can be replaced with ai for most small and indie clients sadly, and that's a big issue because that's where most low-mid level artists got jobs from in the past. We know that not every studio will ask for this kind of level, those are usually the bigger ones, and we know for sure that there aren't that many of them for all of us to get employed.

Fundamental skills are what is necessary to get ahead of the curve, and that is when tools become an asset to speed up production. But with any tool, a lack of skill will be obvious very quickly. I'm not affected by it in my job and adapting to it, there is a lot of time saved in production, these tools are great asset generators (especially when based on the own work) and one has to be aware of the fact, like with Photoshop brushes back in the day, once the tool dictates the outcome it's a sign of being a bad designer and illustrator (and that skill improvement is required). Experience is key indeed. With that said, Feng, glad to see you again and hopefully you're well and healthy.

I can't express how glad I am to see you again, Teacher Feng!!

Great to see you here Feng, these last few years have been a challenging time for all of us. Only just beginning to see the light at the end of the tunnel. So glad to have you back!

Respect the work of concept designers and artists. However, it's important to understand that their focus isn't on solving real-world engineering, industrial, or architectural problems. While there may be some overlap in design principles, their work is primarily for the entertainment industry. The end product is consumed by individuals who may not fully appreciate or be directly impacted by the design beyond an emotional level. The difference between experiencing something in real life, like driving a car, and seeing it on screen or in a game is significant. This disparity, when considered alongside the business aspects of the entertainment industry, is substantial and something that corporations and big tech are keen to minimize. Do you really think the average person would care about the full functionality of a sci-fi vehicle on screen unless they owned one? 😅. I am in no way trying to belittle the job of a concept designer, was just stressing the point that - as long as there is no meaningful impact in the minds of audience with these design choices, nobody actually cares in the long run. While on the other hand engineering or actual industrial design is something which directly impact the quality of life itself, not just for entertainment purposes alone.

Oh man, I have quit pursuing art professionally months ago due to AI. It's just not worth the long years of effort compared to what the AI has been able to generate lately. The decline of societal value in real draftsmanship is just not much of a worthy pursuit any longer. Public/ or business perception of what AI is or what isn't, may not even make a difference going forwards in the years to come. However, I did enjoy my days drawing, it is fun being in the flow state, drawing things you like. I wonder what the future holds when it comes to technology and art. And I wish I was wealthy enough to have the courage to pursue it full-time, but it's just I cannot do it due to how unstable it is career/ market demand wise -- maybe I'd pursue it as a hobby during the weekends.

We really missed Feng! It's great to have you back

Welcome back Feng! I still remember 8 years ago listening to your podcast design cinema, I have learned a lot since then, though still need to learn a lot more. Thanks for what you do with design cinema videos!

His back!!!!🎉🎉🎉

I have said this before, I will say it again in all the chaos that some people feel in. AI is a tool. Not an end result. Period.

I swear listening to your voice after so long it's like I've traveled back in time ! Glad to see you again !

OMG! Thank you so much for this videos and the future ones! I am so happy that you came back! Thank you!

Feng, AI has a lot of things I am not sure you are aware of (and it does not dispute what you are saying here let me get that straight) but there are inbetweens you may not be aware of. Such as dedicated (free) ai models which tbh work better than the paid ones, since they use community made designed models and custom trained things to achieve specific results, and things like inpainting to move things along faster or inject your own art skills in with, and so on. 50:55 This is a good example of what inpainting does well. 1:03:01 This is a good example of how AI may be used in the future. Because a custom trained model would be able to do this to a degree, but it would require the base made by the actual artists and probably have an ai person on staff handling and training it and updating it with their work to provide them a model they are using to generate things to help their regular work flow move faster; most likely using it to brainstorm a few things and see what -kinda- works before dropping several hours into said things. 1:06:38 Inpainting can do that, but unless the model was custom trained it would be loaded with errors and so much would need to be re-done that it would probably be easier for a skilled artist to do it manually. Unless AI becomes incomprehensibly more powerful, I doubt it will ever be able to do this right. Things are going to be interesting at any rate.

I appreciate the video but it makes me sad, to be honest. Maybe I misinterpreted it but: 1. illustrators will lose their jobs, no matter their skill or style (AI can replicate that). 2. entry-level for artists is unbelievably high to reach (it seems that the new artists' generation is pushed to become "AI engineers") And unfortunately, there are already 3D AI models. Seems like a matter of a few years before it synergizes with 2D (Midjourney) and text (ChatGPT) models to create thoughtful designs and storytelling concepts (ChatGPT already works with Stable Diffusion for example). I truly hope that indie studios will arise with "Made by Humans" mentality and enough audience (probably also those replaced by AI?) to use their content.

Design Cinema has been so important to me ever since I was a child. I checked this channel just 2 months ago to see if there was any new videos that Feng had created. I am beyond overjoyed to see this. My night is made.

Wow, I'm so glad for another episode. This series has been one of the most valuable resources over the years. Thank you so much, Feng Zhu!

b.. but.. how do you gain experience in professional field when all smaller studios starts using AI and bigger studios are laying their artists off..? (Riot recently laid their Legends of Runeterra team off)

All the fears related to AI comes down to the understanding of what AI can and cannot do, the more you understand it the less you fear it and realize AI it's just a glorified calculator of words, even the name is misleading because the AI don't have intelligence per say. But as always you approach was 100% on the REAL problem, explaining the limitations of AI is the key to not have any fear about it. Thanks for this video and welcome back boss!

I am so glad you returned! One of my favourite chanels!

this channel gets my art journey started back in 2018 and years later i'm taking commissions. Didn't expect an update for sure but here we are. Couldn't be more happier

Lets gooo

Just wait til FZD starts teaching its students to us AI. Unfortunately it's looking like that's the future because Feng is unaware of what AI can already do. Entey level jobs are being squeezed. Instead of 10 artists production will only need 3 and AI. Then who will pay to go to your school? If AI is what the industry becomes, then AI is what FZD will teach or it will go out of business.

i missed you mr feng, thank you for doing design cinema all these years

Soooooo HAPPY to see you back!!!

AI researcher here that occasionally does concept art as a hobby. I like this video so much, this gives a great summary of what AI can and cannot do currently while quenching the fears of "it will take my job". I've always said about these kind of Diffusion Models (current text-to-image state of the art): adapt or die, AI can make stuff that look pretty but lack substance. However, I would be careful with the assessment of "AI would never be able to do X". These kinds of shortcomings shown in the video are well known within the AI research community and are actively worked on by some of the smartest people on the planet with massive amounts of money thrown at it. It would take a breakthrough or two more, but AI will be able to reason better and better. Current shortcomings of text-to-image models are also present in large language models like GPT4, and I assume if the grounding and reasoning is improved, it will also be applied to text-to-image and text-to-video. BTW, I heard FZD school was closed for good? It isn't clear from the video what will be the future...

Respectfully, I've watched you for years now, and I'm very dissapointed in your analysis on AI. I understand this is a Pep talk, but it has an underlying attitude of accepting AI. While you and your peers might not personally be affected, the people who do illustration, book covers, comics, commissions, etc, are deeply affected. There is less work. People are being offered lower pay. Mass layoffs in creative industries. Your argument is that AI can't currently do x number of things. But to that I ask: what if it could? How would you talk about it differently? To me it does not matter what AI CAN do, but rather what it SHOULD do, and how we reconcile the stolen profits and labour from artists. AI is fundamentally different from Photoshop, ZBrush, Blender, Sketchup, etc. The former is software designed as a tool, using the knowledge of the creators. Regenerative AI on the other hand, is built using billions of images scraped from the web, made by unpaid artists works, and ignoring copyright. If we allow this to continue, it will just keep eating the internet and our outputs to compete with us in the market as an alternative to our labour, USING our labour. The product is both business-to-business and business-to-consumer. It supplants the need for artists in so many industries, by using the artists works against them. This is not moral, and tons of lawsuits are challenging the legality. Art is many things to many people, and not everyone values industrial design. Who benefits from regenerative AI? It is primarily the companies, because artists rates will not reflect positively the "increased productivity". Technology is not manifest destiny. You have a voice, and I wish you would use your voice to denounce the immoral practices of these companies, instead of downplaying their complete disregard and disrespect for personal data and art.

Thanks for continuing to share your wisdom through Design Cinema over the years! Your presence adds so much value, and we’re glad to have you back, Feng!😁

ai doesn't have to replace all the artist, but if one artist can do 2 artist jobs now, then half the jobs is gone. thats the point.

Once in one of your videos you strongly recommended for people to use their own photos to Photobash, and that a person don't have the right to use other people photos. You STRONGLY pointed that. But now you have nothing to say about the millions of artists that are having their work taken as if it didn't have any kind of copyright?! That's contraditory. The artist getting their art sampled are getting NOTHING, and those companies will NEVER compansate all those artist for taking their art, and this is my main problem with ai

Glad to see you back Feng. Just know that every single minute you spend (and have spent) on these episodes is greatly appreciated by many people.

Oh my god! You cannot believe how happy I am to see you back! I just started working on a indie game project with my friends and I had to think so much about all the great lessons you taught us over the years while creating concept art!

Yo FZ, thank you for your valuable input. I would just like to point out that AI models aren't "currently" just out there sourcing from the internet every day. You say a couple times something to the effect of "it goes and gets it from the internet", but that isn't technically true, as these datasets have been compiled and trained a couple years ago now (laion 5b, for example). Some users have created their own finetunes and... loras or wtv they call it which add additional "expansions" to the datasets for more specific styles/elements, but that's all. There is no active scraping and incorporation.

the sad thing with ai is that it removes half of the fun out of the process, designing is fun but drawing is really fun, ai just removes drawing completely

Wow what a great thing to find waiting for me on KZfaq… a new video after such a long long wait… your voice is so familiar it’s helps revisit all the enjoyment I gleaned from your past presentations. Thank you for this . Now on with the show 👍🏼

Thank you once again for put in your time for this amazing content, and for free! 🙏🏽

Art and Design is like a language. AI does not speak that language. It is like a skilled Caucasian comedian speaking gibberish to sound like he is speaking Cantonese. Sounds the same to a non Cantonese speaker. But to a Cantonese speaker, it has 0 meaning despite its similar sound and tones. AI art might be useful for giggles, and surface level, disposable visuals. Editorial imagery probably will find AI art useful for the density of images needed, and the disposable nature of it.

What AI Cannot Do - for now

This video is the real MVP of AI videos for artists. Thank you yet again Feng, so so much love to you!

those future projects sound so exciting! its so nice to have you back and learn with you

I dont think AI can be considered a tool when it can do the whole thing for you... I mean sure, you can use it as tool for productivity, but on the larger scope of things, its essentially you optimizing yourself out of a job. The skill floor and ceiling become a single line, so by the time the actual artist has finished a painting, even with using AI as tool, already has the unskilled used the same amount of time to produce hundreds of images, for they used it as not a tool but automation. The later exceeds in output, vastly. And unfortunately, optimizing production and cutting corners is in the lifeblood of management. "Good enough" comes to be preferable ever more so. So its a big tragedy of the commons, as no one stands to gain anything long term but the owners of this technology. After all, by using AI you are also training AI. And its merely a belief that it'll need you forever. MJ6 has largely automated prompting. So, though displacement is imanent in like, every industry, I can see from here that the endgame of this is direct-to-consumer content generation based on what the algorithm thinks you like. So we need to start thinking on those terms, even if we are overshooting. Is it that surprising though? Suno can generate 3 minute songs in 20 seconds, aka the time to produce is lesser than the time to consume. So it's not crazy to think that as models improve it'll be able to infinitely generate music for you to listen based on your "preferences". Same should be possible for images, video, text and even code...

Mr. Feng Zhu Why are you so early? You should have given yourself another 10 years... Other artists have gone much further in that time. And now everyone is failing due to a lack of jobs because AI is here. The title should be: What AI can't do yet! AI will take over everything sooner or later, and concept art is no exception. So don't get your hopes up. It's no longer worth working as an artist in general.

Can’t tell you how much I’m happy to see you back Feng! Thank you for the detailed update and insights into the industry moving forward. I’m a 3d environment artist for games, but I’ve been following your work for many years since watching your early design DVDs at Gnomon. Thank you for continuing to share your experience and insight!

I'm really glad you're back, but most glad you got a break 💜

Vary region lets you prompt specific areas, so you prompt the general scene, then prompt each detail. Could be said AI can't one-shot these yet but this can be done by AI directed by humans like all AI art. Same is true for character consistency, we can use --cref character references to train the AI, you can also give it --sref to use existing IP as style reference to make new characters to then cref. Some of the meaningful ideas that direct our art, humanity only captured these ideas after thousands of years, so we cant expect AI that was just born, it will grow meaning awareness quickly, but it will take time.

What AI Cannot Do Yet*, fixed it

Thank you, Feng. You've been a great influence for me, not only as an artist, but also ifluenced my work ethic and shaped my worldview in some sort.

Whoa, so good to see you again man!! Thanks for this upload!

Sorry but AI is NOT a tool. And there are NOT plenty of jobs.

So nice to see you back Feng!

so glad you guys are back. Love your content, you guys helped my art grow so much

It's so hard to overstate how good it is to see you again, Feng! It's like seeing an old friend from decades ago coming back online. Welcome back!

Been watching since 2011 and have learned so much from you over the years. It's great to have you back!