The answer is:
These days no one will retire or hang up the spurs in the career they had as they came out of school. You will have an average of 2-4 careers in your lifetime. Embrace it.
I’m on #3.
The answer is almost always in the middle somewhere.
Roto is not done. Foundry and DNEG and the University of Bath killed their SmartROTO 4 years ago. If there was a company that cared enough about eliminating roto, it was DNEG. If there was a company that had the firepower and cared enough to sell software that eliminated roto, it was Foundry. They gave up 4 years ago. DNEG donated something like half a million artist-provided shapes, and over 100 million keyframes to build the roto models of all models. And guess what? They gave up.
A short history with sources, cuz that’s how I roll.
Bath announces…
https://researchportal.bath.ac.uk/en/projects/smartroto
Foundry announces…
“We gave up. Here’s why…”
Roto has changed in the last 5 years, for sure. Sure, sometimes RunwayML is good enough for some segmentation for instances that do not require pixel-perfect detail that we require sometimes. Sure, AE’s Roto Brush means shitty AE compers can avoid costly roto sometimes for whatever it is they do. Sure, Resolve’s Magic Mask helps colorists reduce the number of times they have to fart out multiple passes for that tasty roast beef sandwich bite and smile so they roast beef is the right color and doesn’t take away from Arby’s brown. Sure, sammie-roto and matanyone can be good enough sometimes to do certain things. But roto is a tool, a necessary tool. And it and its unromantic but entirely necessary existence will not die.
American’s still write checks, for jeezuz sake. We will buy cars powered by liquid death. The fastest way to send terabytes of data across the world is still FedEx. Record albums, 8 Tracks**, coin-operated laundry, pagers in hospitals, fax machines in medical and legal settings…they…still..exist..and..serve..a…purpose.
Regarding the future? CG ain’t the future. Never was. It was only a viable business for a short amount of time, and that time has long since past. CG is the rotisserie chicken of the VFX world. We literally lose money on it, hoping you’ll pickup a bottle of California Cab at a 250% markup to go with that cute Coq-au-vin recipe you saw in the Times and thought you had time for but settled on the rotisserie chicken cuz life. CG is a loss leader, designed only to increase the access to the kinds of work that are still high enough margin…editorial with creative fees, Flame, color, and design.
CG exists because there are times we want to see something that doesn’t exist. And it’s too expensive to invent it, build it, or heck, even. photograph it in our own back yard.
You literally do not know if an AI tool is going to be good enough to solve a problem until after you’ve drained 17 freshwater lakes in rural Iowa to only chuck out something that barely reaches Slay on the Rudiger’s Continuum of Artificial Forms.*
So, you fire up Maya and make it work. Or, if you are only an AI Artist, and void of all traditional VFX skills, you change your Twitter handle and fuck off cuz that’s the only way your client found you and communicates with you cuz you posted someone changing clothing on Twitter and thought that’d be a swell person to hire to change your celebrity’s uniform from Team 12 to Team 13. Dangerously close to a true story.
My dream is for AI to bring the CG studios back to their glory days. I mean, isn’t it just a branding problem? THEY ARE LITTERALLY CALLED COMPUTER-GENERATED STUDIOS! AND THEY ARE BEING TAKEN OVER BY AI? ARTIICIAL INTELLIGENCE? AI IS CG!
No more CG. No more AI. Call it CI. Custom Images. There. I fixed the problem. Oh wait..if you’re British you’ll want to call it BI…Bespoke Images. CI for North America, BI for Western Europe.
*Okay I lied. 8 Tracks died. They actually died. RIP 8 Tracks. 1964-1983.
**RUDIGER’S CONTIUUM OF ARTIFICIAL FORMS
shit → slay → specific → strategic → sellable → showpiece → signature
True
You just wait for buttonification (the turning of these vastly over complicated systems into simple widgets). I think it’s that simple. Spending a bunch of time on GitHub repositories installing ridiculously specific and difficult to learn homebrewed interfaces is not the way in this environment. That’s a waste of time. The vibe is words to images. You’d be better off reading Proust and Melville than learning comfyUI in my opinion.
Your window of offering something in terms of creating that a 14 year old can’t is exceedingly small as things like veo come into existence. So thinking, I’ll offer this and no one else can! is flawed thinking in my opinion. Maybe for a blink you will, but not for long. Things will still come down to eye for quality and taste (a large part of the compositing as is) so you can still lean on that, but thinking that by learning some interface that’s really hard to learn and hacking it all together in some technically difficult way while buttonification looms and blah blah, is just setting yourself up for disappointment when the agencies internals just click the go button.
This. My secret pleasure is that behind the scenes, people who know how to effectively communicate complex concepts via the written word, have a vast knowledge of disciplines, ideologies and modalities and can effectively weave together said language with those references, are ironically those with the most to gain in this new visual space.
I almost included Claire Keegan in there haha! Literary fiction is the sleeper skill-up
I think you just have to get in the habit of not getting too busy with just one thing.
Freelancers have the same problem with clients. You spend time networking and pitching. Then you get a big job that takes a few weeks. You’re heads down. When you come up for air, ready for more work, you realize you have neglected your pipeline, made no calls, sent no emails for a few weeks. You have to find a way to multi-task. Take 30-45min at the beginning of each day to feed the other parts of the business, or your curiosity.
And you’re never to old to learn something new. Sticking with it to the end assumes stability, and as pointed out, that doesn’t exist. There’s always time to learn something new.
But @BrittCiampa is right - there are tools and then there are methods, and the ability to see the bigger picture. Tools are interchangeable and short lived. Methods and understanding visual media and story telling at a higher level has much more longevity. Don’t focus on the tools, master the the craft.
A story Andy likes to tell (and it’s a variation of a story that exists in many forms and crafts): A couple sit at a restaurant across from Picasso. He paints something on the napkin, they ask for the price and he says $50K. They are shocked “But you took 5min to draw this on a napkin”, Picasso responds: “Yes, but it took me 40 years to learn what I just did”.
It’s not about the napkin or the pen he used. Those are tools. Understanding how to reduce a whole scene to a few recognizable and meaningful pen strokes was his craft that has value and endures. He could paint that same thing in a prison cell with his blood after pricking his finger, and still achieve the same result.
Flame is a tool, albeit an incredibly good one. What we all use it for to make good images greater, is the craft.
If there is no more Flame, there may still be Nuke. If there are none of them, maybe there are new AI tools that can do it. You still need to prompt them correctly.
You can copyright methods, but not tools. Tools you can trademark.
I would argue that ComfUI is already buttonified to a large extent and vastly less technically difficult than say learning Flame from scratch.
I would be very very surprised if pure txt2img prompting ever generates a watchable film or TV show. You can see it in all the Veo stuff. This is clearly the biggest step forward in the last two years. Realism overall is vastly improved, but most of the large structural issues remain unsolved.
Modality guidance is the way forward at least in the short term. ComfyUI may not ultimately be the tool that everyone uses, but neither will it be an empty text box and a big red generate button. At least I hope.
That’s just buttonification then. What are you to do? Learn the buttons if you want to use it. Done. I don’t see how generative ai gets past language (the image is language by the way in many schools of semiotics, in many schools of philosophy, everything is just language). Gotta communicate what you want and even if it’s image to image, you’re still going to have to communicate and written and spoken word are our means of communication.
Should have. Her books are just slightly longer than the prompts you need to keep your sdxl images from looking like they were rendered in 32 pack Crayola.
I asked Microsoft copilot to create arguments for hiring an expert digital visual effects artist over machine learning tools and the counter argument.
Hiring a Seasoned VFX Artist:
Machine Learning VFX Tools:
TLDR - blah blah blah - wordy blah blah - second argument - point 1 - CHEAPER - WIN
I’d argue that in terms of monetization of this technology (paid user base coupled with time on device enabling maximum data rendition for sales to third parties) necessitates a text box with a red button that says go. The goal isn’t to provide a helpful service in this new attention economy, it’s to capture minutes of your life so they can be plopped in the big data hopper. And that means everyone needs to use it. Not because everyone needs it, but because it needs everyone to make money.
@BrittCiampa - yep
I don’t really disagree with any of that. I guess my main point of contention is with the idea that reading Proust is more helpful than learning ComfyUI and that txt2img is the only way forward.
This may be a finer point of distinction than is worth making, but images aren’t actually language. Images are images and language is language and the thread between them is that they’re both systems of signs with common signifieds. Humans translate between these two systems through the bridge of signifieds. AI doesn’t use signifieds in the traditional sense. It moves between two signifiers through latent representations. While superficially these may appear to be very similar pathways, they are fundamentally different.
Understanding how your signifiers are likely to be translated into latent representations is going to improve the quality of your prompts more than being generally erudite.
Leaving ComyUI aside for the moment, if you really thinking that single-prompt generation is the future then you’d be far better served by practicing drawing than you would by reading more books.
Ummm, I may need one of these.
Babel Fish - The Oddest Thing In The Universe
Dig! I hear you here (although what you’re saying is from a very specific viewpoint regarding semiotics that we don’t need to rabbit hole, but if we ever grab a drink- I’m super down LOl). I’ll add that I mainly think learning comfyUI is a waste of time because adobe and google are on the hunt and they’re obviously going to win. Might as well just wait for their buttons. But I could be wrong!
Nick Clegg: Artists’ demands over copyright are unworkable
some clever wag should send this clown’s manuscript to chatgpt.
heck, if hs track record is anything to go by this clown’s manuscript was from chatgpt.
we have names for humans like this.
I am not going to say I understand the discussion completely (too much semiotics and media theory for me), but I want to add that images while not language have a story-telling structure to them, especially in our medium of film. There is the book called “Grammar of film language” which AI has to understand in order to tell meaningful stories.
I’d so like to join you for drinks or button-pushing.
Hello!
From my point of view, AI is a great tool that pushes us to reinvent ourselves and why not learn some new tools if the need arises.
I work with a lot of agencies, production companies, and post-production houses.
If we move toward the utopia where AI replaces us, then it’s clear none of the people I just mentioned would still be around.
Clients come to us because they want someone who takes care of the image, someone who can reassure them, someone who knows how to tell if an idea is good or even bring new ones. An expert.
And all those people don’t want to lose their jobs either.
AI has no soul it works because social media created algorithms that hack into people’s brains.
Personally, I don’t see any beauty in it, just a tool to help make beautiful things.
It will allow people with no budget to make low-cost ads.
But they’ll still dream of making a real one, with real people, to live the full experience.
I do agree that, in the long run, jobs might decrease because of this but there will always be some left.
Semi-related first hand experience on how life with AI is.
After having a well documented somewhat critical view of AI, I had just stared warming up to the performance of Claude after some time saving and well done experiences.
Yet, yesterday I had a long chat with Claude, which at the surface delivered good results, and in fact I initially missed its hallucination until someone prompted me to double check. Sure enough it had substantial math errors. I went back and forth with it at least 6 times, during which every time it acknowledged the error in polite language, promised to do better, only to screw it up again differently.
At the end I gave up, it apologized. I asked it for a refund, and it sympathized with me, but said as an AI agent it couldn’t authorize refunds. So we left it at that.
So I’m back to picking up the data points from its answer that were helpful and doing the result manually.
And no, this is not a joke.
We’re 2+ years into this endeavor. While AI has showed some impressive results, it has also demonstrated that it’s not a replacement for human thinking for quite a while.
It certainly gives the appearance, and less careful users may be tricked. But it remains an eager but unreliable entry level employee for the time being, who will not be promoted at review time. Management will be disappointed and forced to regroup.
Given that Flame artists generally work in a high impact / high stakes environment, we will see a few car crashes in slow-mo, but we will live to work another day.