Schadenfreude or Sadness - The Slow Death of Artistic AI Tools

[ghettosynth]

As the pressure of investments needing justification and the inability to find profitable applications continues to mount, we start to see the first victims. Laugh or cry, or discuss, your choice.

https://variety.com/2026/digital/news/o ... 236698277/

OpenAI said it will discontinue Sora, the generative-AI video creation platform it launched in late 2024, without providing a reason for the decision.

“We’re saying goodbye to Sora. To everyone who created with Sora, shared it, and built community around it: thank you. What you made with Sora mattered, and we know this news is disappointing,” OpenAI’s Sora team said in a statement Tuesday.

I think this is the big news related to this however:

Disney Drops Plans for $1 Billion Investment

Disney, which had inked a major pact for Sora, said it 'will continue to engage with AI platforms to find new ways to meet fans where they are'

I think that this is still far too expensive and just not ready for prime time. If you want to see a Sora video, you know that you've come to the right place for genuine AI slop, here you go. This was created with three 10s scenes stitched together without fanfare. Google's audio generator created the jingle.

[Dirtgrain]

Definitely slop, but not it's got me thinking, would Botox improve my arse?

[BertKoor]

The cracks are showing everywhere.

Screenshot_20260325_000038_Mastodon.jpg

Hold tight for the bubble to burst!

[ghettosynth]

Karpathy is correct here. I agree 100%, but, I still use LLMs to code for hours every day. If that seems like a contradiction to you, then you aren't experienced enough to use LLMs for coding.

[VariKusBrainZ]

AI doesnt even know the current thing

[BertKoor]

[...] , then you aren't experienced enough to use LLMs for coding.

In other words, it brings benefits to experienced developers only and is dangerous to use for novices.
Not only is that a disguised form of gatekeeping, but it does not help the industry in a sustainable way. And that is true for all forms of sustainability.

Lemmings, anyone??

[ghettosynth]

[...] , then you aren't experienced enough to use LLMs for coding.

In other words, it brings benefits to experienced developers only and is dangerous to use for novices.
Not only is that a disguised form of gatekeeping, but it does not help the industry in a sustainable way. And that is true for all forms of sustainability.

Lemmings, anyone??

Only if you see what I said as a contradiction. You seem to be attached to projecting gatekeeping. Who is it that you think should help "the industry" here in a sustainable way? And which "industry" are you referring to?

I have repeatedly stated that LLMs are incredibly useful if you know how to use them. If you don't know how to use them and you would like to, there are plenty of resources to help you with that, e.g., ask an LLM.

I'm not sure what your focus here is? I agree that many current attempts at deployment are not sustainable. One of the key ways that I think that this is important is via what is refereed to as LLM fatigue, or, the increased cognitive load from recreating intent from scratch in order to moderate LLM edits to a codebase. Not everyone is operating that blindly however.

So, if you want to discuss, then discuss, but vague pejoratives are not helpful.

[BertKoor]

Who is it that you think should help "the industry" here in a sustainable way? And which "industry" are you referring to?

The industry I'm referring to is Software Development in general. One of the characteristics is that there has been historically a nearly exponential growth in the number of software developers. The median years of experience is (and has been for decades) five. Today we have about twice as much developers as five years ago. So in average only half of my colleagues do this job for more than five years.

Such rapid growth comes, as you'd expect anywhere, with specific problems.

Firstly, novices need training and gather experience in order to become experts. You may have heard of the "10.000 hours -> expert" rule? With 2.000 working hours in a year (40 hours a week, 50 weeks per year) that's five years. Can we maybe use AI as a teacher or mentor? I have doubts, although it may help a bit.

Secondly, not everybody has the skills and talent, is cut out of the right wood to be a programmer. It has, in my experience, always been the case that the End Users of software found it difficult to get what they specifically want. Their dream has always been that they could create software themselves without a programmer as an intermediate. I've seen and used plenty tools that looked promising in making software development easier. Without exception these all started out smooth, trivial software and simple prototypes could surely be built with it. But there always was a point where as requirements and complexity grew, the tool (and or its users) had painted themselves in a corner and it just could not deliver anymore.

It has been the premise of Generative AI that it can replace the programmer. Just state your requirements and you get your software. It turns out it's not so simple, you even need to be an expert to get the most out of it. How does that help any of us?

[BONES]

In other words, it brings benefits to experienced developers only and is dangerous to use for novices.

That might be a reasonable assessment if this technology was well established but it's not, so everyone is still really a novice. Nobody has a 20 year head-start so it is still a pretty level playing field.

Not only is that a disguised form of gatekeeping, but it does not help the industry in a sustainable way. And that is true for all forms of sustainability.

So what? Nobody has a responsibility to maintain an industry if it is made obsolete by new technology. That's Luddite thinking.

Old jobs go, new jobs replace them. I've seen it twice in my industry in the last 30 years. When I first started, we still made corporate presentations with 35mm slides and if we wanted to use a stock image, we had a rostrum camera to take a photo of it from an expesnive, gloosy hard-bound book. Then we had the era of Qantel domination, the first wave of computer-based graphic design. They were horrendously expensive turnkey systems and the artists who used them had incredible skill and talent but eventually they were squeezed out by desktop computers so all those amazing people had to retrain or look for new jobs in another industry.

In parallel to this was the rise of the internet, which has probably created twice as many jobs for graphic artists as have been lost to the march of technology over the years. And these days those jobs offer better careers and more money than the traditional jobs in print, TV and film so the old fogeys like me, who don't want to embrace the change, are being left behind. (In my case it's more that I don't need to embrace it because I can retire any time I feel like it.)

It's all swings and roundabouts, graphic design is the third completely different career I've had in my 50 years in the workforce - 10 years in the Army, 10 years as a retail manager and now 30 as a designer. I think that's just the reality of modern life. You need to be adaptable and unafraid of change. Life should be an adventure, shouldn't it?

Lemmings, anyone??

It is well known that the common beliefs around lemming behaviour are vastly over-stated.

[ghettosynth]

It has been the premise of Generative AI that it can replace the programmer. Just state your requirements and you get your software. It turns out it's not so simple, you even need to be an expert to get the most out of it. How does that help any of us?

Wrong framing.

The electronics industry went through this exact transition when PCBs and automated assembly replaced hand-wiring. Nobody asked how automation helped the hand solderers, frankly, it didn't, and that was the point. What it did was eliminate the manual layer entirely and push the remaining workforce into design and systems work, that is, if they could handle it. The technician shortage of the period fueled the NRI/CIE correspondence courses, which were the programming boot camps of the era. Software is following the same path. AI is not likely to replace the engineer any time soon, it replaces the assembly work that we spent twenty years decomposing into junior roles and calling 'development.' The people who are threatened, for now, are the ones whose entire skill set is framework gluing and CRUD endpoints, work that AI does at near zero cost. The people who aren't threatened are the ones who know when to apply a design pattern, not just how. That distinction was always there, we just papered over it with agile adjacent processes and sprint velocity metrics. The real shift isn't AI replacing programmers, it's AI exposing that a large portion of what we called programming was never engineering in the first place.

This is Bob Martin's point in a nutshell, the source of your perspective. His entire framework assumes the exponential continues and the solution is disciplining the growing army. He never considered that the army itself might be an artifact of insufficient tooling. The growth wasn't a law of nature, it was a consequence of software being labor intensive in a specific way that was historically contingent, just like hand wiring and component level repair was.

When this shifts up, and it likely will, how far will be the relevant question. Data may be the limit, or it may not. The skill is recognizing the line in real time and leveraging skills on the correct side of that line. This is, hopefully, what one learned to do as an engineer.

Whatever industry that you are in, it's incumbent on you to find where this sharp divide lies, if it does at all. So, if "us" is you, then the question is one of reflection. How can you use AI to help you? If you can't answer that, then you are one of the folks that may need to retool yourself into the new world. Good luck.

[zerocrossing]

Definitely slop, but not it's got me thinking, would Botox improve my arse?

It's spelled "buttocks."

[zerocrossing]

I frankly find it all fascinating. I mean, in some ways it can do some interesting things, and quickly, but in most cases, it has a "mind of its own." If you're just brainstorming, it's fine. Might get a good idea, but really no different than going through internet image searches, or going to a library and looking through art books. If you want something specific, you're SOOL. I've seen a lot of Sora videos where the character is never the same from scene to scene. Kind of the same, but not quite. Sometimes, quite different. Different enough to think, "is this a new character?" Why? Because it just doesn't know

I did hear that Sony (Columbia films) had developed a sort of "reverse AI lookup" that could pin point what the source leaning material was from any given image. I bet a team of lawyers was gearing up for a full on assault, and OpenAI decided to fold.

[pdxindy]

I've seen a lot of Sora videos where the character is never the same from scene to scene. Kind of the same, but not quite. Sometimes, quite different. Different enough to think, "is this a new character?" Why? Because it just doesn't know

It's improving fast, both for character consistency and scene consistency. There's new storyboarding tools and new models.

[zerocrossing]

I've seen a lot of Sora videos where the character is never the same from scene to scene. Kind of the same, but not quite. Sometimes, quite different. Different enough to think, "is this a new character?" Why? Because it just doesn't know

It's improving fast, both for character consistency and scene consistency. There's new storyboarding tools and new models.

But they shut it down. At least this version.

I’m actually thinking about using something similar to make a little music video. I had an idea that would be impossible for me to achieve without this technology. It will have to have a story board function, but character consistency isn’t critical. I think it will provide a result that is as good as I need it to be, but not what I’d expect from a professional production.

I think we’ll still see a lot of slop from casual users, but I also think that we might see some good stuff too, especially when it’s coming from people who understand film making. I saw someone post something on LinkedIn last week that was ponderous as hell. Each cut was OK, but the pacing was far too slow, and there was no sense of narrative arc. Just a noob saying, “ooo! That shot looks cool! Let’s do another… and another.” Slop.

[pdxindy]

It's improving fast, both for character consistency and scene consistency. There's new storyboarding tools and new models.

But they shut it down. At least this version.

Sorry, wasn't referring specifically to Sora. There are other tools coming out. Martini looks promising for example. And Seedance has also improved consistency.