AI Art’s Hollow Victory: Faster, Cheaper, and Forgettable

We aren't AI artists

Image made with DALL-E 3

About two years ago, we began seeing the first significant advancements in AI-generated images. These images started to look noticeably better. The strange shapes and unnatural features were being replaced by more realistic ones. Still, I couldn’t shake the concern that, even with these improvements, I wouldn’t like them. At that point, I started questioning not just the fact that this new art was being created using the work of artists without their permission or compensation, but also the very nature of AI-generated art itself. It felt like it devalued human creativity, existing as nothing more than a cheap imitation.

While a few years ago that critique may have been true, it was clear that the technology was steadily improving as generative models advanced. I speculated that, within a few decades, once this technology was fully refined, it would lead to a major—and possibly permanent—shift in how future generations create art.

Imagine a world where people no longer need to learn how to write songs, play instruments, draw, or make movies. They wouldn’t need to understand the finer details of what makes great art. Instead, they’d rely on simple prompts and AI systems trained on the collective body of human-made works. Why would anyone spend tens of thousands of hours mastering an art form when they could just go to a website and generate something similar instantly?

It feels like this could be the beginning of the end for the arts.
I couldn’t help but feel that AI-generated art wouldn’t just ruin careers but could also destroy art as a hobby. It felt like the start of a dystopia.

But is AI-generated art truly inevitable?

It’s obvious that generative AI has continued to improve, and the arts haven’t been exempt from this trend. In fact, its progress has been much faster than I ever expected.

Right now, the internet is flooded with AI-generated content—some of it decent, much of it bad, and some outright terrible. Entire industries have gone on strike, worried that this technology will replace their jobs. Meanwhile, companies like NVIDIA have seen their market capitalization soar to trillions of dollars, fueled by the AI hype machine. In my view, this is a serious issue, signaling a bubble that keeps getting bigger.

And yet, despite all of this—despite all the advancements in AI technology—it seems to me (and probably to you as well) that the only real, practical use for AI-generated art remains the same as it was at the start: not particularly meaningful or impactful.

Several companies have tried creating ads using AI, but the results often fall short. The ads tend to look overly artificial and far from natural. While AI-generated images have undoubtedly improved, we’ve also gotten much better at recognizing them for what they are. This makes sense because removing the human element from the creative process leaves AI-generated images feeling recycled or generic—just the product of a mathematical algorithm.

The Art of the Absurd

In early 2024, OpenAI launched its video generator, Sora. While the demos OpenAI shared weren’t perfect, they represented a significant step forward compared to other AI video generators available at the time. Instead of just creating static images, Sora could generate short films, brief clips, or at least a few seconds of semi-realistic movement.

Now that the initial hype around AI has settled, it seems most people aren’t as impressed with Sora as they were a few months ago. This is partly because other companies are offering similar tools, but also because OpenAI’s initial presentation only showcased the best-case scenarios of Sora’s capabilities. In practice, trying to generate a typical video often results in something that looks almost real but doesn’t quite get there. That’s not surprising, though, since Sora likely used the entire internet—including copyrighted material—to train its models.

One notable incident involved an OpenAI researcher accusing the company of using copyrighted data. Interestingly, the researcher came forward right before being summoned to testify in court, where they planned to present evidence. If proven true, this could have caused major legal issues for OpenAI, potentially even threatening the company’s existence by exposing it as essentially stealing from creators. Unfortunately, we’ll likely never know the full story, given the researcher’s untimely death.

As it stands, Sora is still an unfinished tool, frequently generating results far removed from what most people would expect. However, this alone isn’t enough to dismiss AI art as the future. The technology will continue to improve, and as it does, its success rate will increase. To be fair, Sora is a decent tool, flaws and all. But I believe its shortcomings, especially when applied to art, go deeper—they’re rooted in the very concept of AI art itself.

I’m still skeptical about AI’s ability to create complete works of art solely from text prompts. It’s not truly creating art; it’s pulling from the data it was trained on, which inevitably includes human-created works. The output is often generic. Supporters of AI art argue this isn’t a problem, claiming that all human art is also inspired by earlier works and that AI is simply doing the same with its training data. But I disagree.

Whether or not you believe that all human art is derivative, there’s no denying that somewhere along the line, a spark of originality occurred. AI, as it stands, isn’t capable of that. Its understanding is entirely based on existing works, and it lacks the human touch that brings originality and meaning to art.

I’ve seen people try to make short films with generative AI, but in my view, they never quite succeed. Take the shorts produced by TCL Studios, for example. These five-minute videos feel more like compilations of clips that are clearly AI-generated. There’s no consistency in artistic style; it’s as if each film were directed by someone completely different. The actors change, and the stories lack any real cohesion.

Control and Consistency: The Defining Challenges

Control and consistency, by their very nature, conflict with the goals of generative AI art. The primary aim of AI is to make creating images and videos easier, but it sacrifices control in the process. Even if tools for video editing help address this issue, the challenge of maintaining consistency remains. Consistency is what allows creators to reuse assets and control keyframe movements for characters on screen. To truly solve this problem, we’d need data points in a 3D space—essentially, CGI.

This isn’t entirely out of reach for AI; we’re already seeing the beginnings of progress in this area. But the more control you introduce into a program, the less time you save. Even in ideal scenarios, the inconsistencies and errors in generative AI models would still need to be corrected by skilled professionals. When you compare the time spent fixing these issues to doing the work without AI, the savings quickly diminish. And that’s the fundamental problem.

For perspective, large language models (LLMs) require massive amounts of data and significant financial backing to operate. There’s a reason OpenAI was nearly bankrupt before Microsoft stepped in with further investments. The reality is that AI still isn’t profitable on its own. It’s also why OpenAI now charges $200 per month for its latest model.

I remain skeptical about AI’s future. I think that as soon as AI starts losing its appeal with the public or stops advancing as quickly, OpenAI will shift its narrative to claim new breakthroughs—perhaps touting progress toward AGI (Artificial General Intelligence). However, as AI becomes more expensive and the hype continues to fade, investors will demand that these tools begin generating real profits. If costs rise and the main selling point of generative AI is saving time and money, what’s the point if humans have to spend even more time fixing its mistakes?

I believe generative AI will have a place in art, but its role will be limited. I don’t see it being ideal for projects like short films or TV shows. It’s not that I’m against AI altogether, but the issue of using training data without permission from artists is something that must be addressed.

That said, when I look at the quality of AI-generated content, it’s clear that most of it will end up as stock images or videos—forgettable and uninspired. Silicon Valley seems blind to the fact that the more progress we make in this field, the clearer it becomes that AI-generated art isn’t as valuable or useful as they make it out to be.

Final Thoughts

Maybe I’m underestimating all of this. Maybe I’ll end up being wrong. After all this time, perhaps you’ve seen clearer benefits of AI in art that I’ve missed over the past two years—or maybe it really has just been another empty promise.

I’m not convinced that companies like OpenAI are adding genuine value to the art world. Typing a few words and generating a short video feels more like a novelty than anything meaningful. That doesn’t mean similar technologies couldn’t have an impact. For example, NVIDIA has shown the ability to create 3D videos from text prompts. This is fascinating because producing that kind of work traditionally requires significant time and effort, like capturing an object from multiple angles. Their AI can even generate polygons compatible with traditional 3D CGI workflows.

So yes, there are ways AI can bring value to art—but probably not in the way many companies are currently pursuing.

What worries me—and many others—is that these technologies might become “just good enough” to work without human involvement. Even if the results are generic, people might still accept them as entertainment.

AI can speed up the process of creating art, as we’ve seen before during the rise of digital tools. But if we try to strip away all the skill and time from the artistic process, we’ll end up with something that, at its core, isn’t really art anymore.