The greatest movies never made

A look into the world of AI-generated movie trailer slop.

While procrastinating on YouTube, I saw a video thumbnail that successfully grabbed my attention. It wasn't a shocked Mr. Beast-style expression or a clickbait-y title. It was a movie trailer for Iron Man 4.

I've done a pretty good job of keeping up with the Marvel Cinematic Universe, and I was a little shocked that somehow I had missed all of the announcements of a fourth Iron Man movie. Spoiler warning: I clearly remember Tony Stark dying at the end of Avengers: Endgame.

So, I watched it.

The good news was I had not, in fact, missed an Iron Man 4 announcement. The bad news is I forever lost two minutes of my life watching an AI-generated movie trailer.

As it turns out, there's a growing trend of AI-powered fake movie trailers on YouTube, from obvious slop to edits with surprisingly high production value. What’s more, these videos are getting millions of views - but why were they being made? And who was watching them?

The Good, the Bad and the AI

The first thing I did after realizing the truth was to see what else was out there. And after only a minute of cursory Googling, I found multiple channels with obvious AI thumbnails. These weren't small channels by any means - the largest had nearly half a million subscribers, and their top dozen videos alone have garnered over 60 million views.

But even between the different channels, there were multiple approaches to content creation¹. At one end, we have the "trailer" for a live-action Simpsons movie. The 2-minute video is a pretty slapdash effort: an AI-generated Adam Sandler voice narrates the video as Homer Simpson, providing uninspiring commentary as still frames are loosely animated.

In case you haven’t been paying attention, the tech to do this has actually existed for a short while now:

• Create a still image with Midjourney, Flux or DALL-E

• Use an image-to-video tool to create a 2-4 second video clip

• Repeat for each scene in the trailer

• Find a celebrity sound-alike voice on ElevenLabs (or another AI speech generator), and use ChatGPT to write a plausible voiceover script

• Generate background music with Suno or Udio for the trailer

• And stitch them all together in iMovie or Premiere Pro

If you're familiar with the AI tools involved, it's not complicated (and if you’re not,

Why Try AI

has a solid tutorial on how to do it yourself).

The Iron Man 4 trailer, in contrast, was a fairly impressive work of video editing. For starters, the "trailer" actually told a compelling story: decades after Tony Stark's death, his daughter discovers an AI he left behind to train her as a superhero.

Perhaps more impressively, though, it stitched together multiple video sources to create the final product: clips from existing films, new (non-AI) animations, scene editing (like recoloring existing Iron Man film shots to create a pink Iron Man variant), and only a handful of AI shots sprinkled in.

All together, it’s actually kind of entertaining! And it made me realize that these AI-generated movie trailers are just the latest iteration of a hobby that's been around for ages: fan edits.

Raiders of the Lost Art

Well before generative AI existed, movie fans were busy creating "what if" trailers for films and castings they wished existed (and before that, fan fiction communities have done this with written stories for decades). For nearly a decade, channels like Smasher, Teaser PRO, and SLUURP found clicks and views by crafting theoretical trailers. They'd take footage of "fan favorite" actors from various films, carefully edit it together, and create convincing previews for movies that existed only in their imaginations.

As for why they do it, an interview with some of these creators sheds some light:

Do the creators actually want to fool people? “In the early days, I was admittedly irritated when people would comment “FAKE!” I would genuinely put a lot of effort into creating these concepts. But it doesn’t bother me anymore,” says [the creator of SLUURP], who is clear that he’s not aiming to punk viewers.

“Some creators in our niche label their videos as ‘Official Trailer’ or omit the word ‘Concept’, which I find can be misleading. It’s easy to see why viewers might feel deceived,” he continues. [Teaser PRO] and [Smasher] agree, emphasising the importance of making it clear that it’s not the real thing; though many people still don’t notice. They also all note that, because of the need to use existing clips through YouTube’s fair use policy, it’s hard to make a profit.

But clearly, AI is having an impact on this process. Instead of spending hours searching for the perfect clip or learning complex After Effects techniques, creators can just describe what they want to see. The YouTubers admit as much:

Fake trailer creators are experimenting with [AI's] endless possibilities. [Smasher], for example, is using it to create thumbs-uppable thumbnails and [Teaser PRO] is utilising it to enhance video quality up to 4K, while [SLUURP] is using AI to speed up the editing process. “It allows for intricate audio isolation and character extraction, which has been a game-changer,” he says.

This accessibility is key to understanding the explosion of AI trailer content. When fan trailers required serious editing chops, they remained a niche hobby. But AI has lowered the barrier to entry so much that anyone with a creative idea can participate. It's the same democratization we saw with Instagram filters and photos or TikTok's editing tools and short-form videos.

Interestingly, I seem to be way behind on this - people have been complaining about the flood of fake movie trailers on YouTube for the better part of a year:

That said, an important question remains: should I be worried about the proliferation of AI-generated movie trailers?

Elsagate

To understand where AI-generated trailers might be heading, we need to look at another YouTube phenomenon: Elsagate.

Back in 2017, parents (and media outlets) began noticing something strange in their children's YouTube feeds. Videos featuring popular characters like Spider-Man and Elsa were racking up millions of views, but something was… off. The content was bizarre, sometimes graphic, and mass-produced to game YouTube's recommendation algorithm.

Source: Wikipedia

The channels behind these videos had discovered a formula: combine popular characters, use algorithm-friendly keywords like "learn colors" and "nursery rhymes," and produce content at a massive scale. The videos were lucrative, pulling in significant ad revenue despite (or perhaps due to) their low production quality. Eventually, YouTube cracked down on these videos after media attention and public outcry.

But even now, the incentives haven't changed. YouTube's algorithm still rewards engagement, regardless of content quality. The platform still struggles with moderation at scale. And viewers – especially younger ones – still click on familiar characters and concepts.

This isn't to say that all AI-generated trailers are problematic. Many are creative, entertaining, and clearly labeled as fan content. But the combination of algorithmic incentives and AI's capacity for mass production creates an environment ripe for exploitation. Just as Elsagate videos evolved to target children's search patterns, I'm guessing we'll see AI-generated content become increasingly optimized for maximum algorithmic engagement rather than creative expression².

Through the Looking Glass

And, of course, this all continues to accelerate. Last month saw the release of Sora, OpenAI's long-awaited text-to-video model. While some have criticized Sora as being underwhelming (in the months since it was first demoed, competitors like Kling and Runway have done a great job of catching up), you can't deny that by being an OpenAI product, Sora will get millions more users than any of its upstart competitors.

Sora promises cinema-quality video generation from simple text descriptions, plus several editing tools to quickly trim, loop, extend, and combine video clips. The barrier to entry for AI video slop isn't just getting lower – it's practically vanishing. How, then, do platforms and users deal with the upcoming flood of AI-generated videos?

The incoming Trump administration is expected to take a lighter approach to AI regulation, meaning legally required safeguards on AI misinformation seem unlikely. Instead, we may see some loose attempts at self-policing, like the watermarks on videos generated by Runway and Sora³.

Some social platforms (Meta, YouTube, and TikTok) ostensibly show labels on AI-generated content. But even if I trusted the platforms to have the best intentions (I don’t), it’s a near-impossible balancing act. How do you distinguish between creative fan works and misleading content? When does a trailer cross the line from "fan creation" to "deceptive marketing"?

The platforms' current solution – pushing creators to label AI-generated content – feels like a bandaid at best (and useless at worst). Detection technology remains unreliable, and creators already are at the mercy of weaponized DMCA takedowns and arbitrary account bans - adding “your content was detected as AI” as a new failure mode doesn’t seem particularly positive. Besides, determined bad actors will always find ways around detection systems.

The Never-Ending Story

Yet perhaps we're focused on the wrong thing. The rise of AI-generated trailers isn't really about technology – it's about our fundamental desire to tell stories, and to see new twists on familiar characters. Fan edits, fan fiction, and AI trailers all spring from the same creative impulse: "What if?"

What if we could generate not just movie trailers but entire movies on demand? What if our favorite shows and series never had to end? What if all of our media was personalized and rendered in real-time?

With AI, we're shrinking the gap between imagination and realization to almost nothing. That can be a blessing or a curse, depending on your perspective. Some will wish for a backlash - a pendulum swing back towards verified, human-created content. Others will accept AI generation as another tool in the creative toolkit, just as they accepted CGI in movies or autotune in music.

I’m actually not sure whether AI video generation should be treated much differently from CGI, at least not yet. The real challenge - for platforms, creators, and audiences alike - is finding a way to allow creative expression to flourish, while ensuring our storytelling remains meaningful.

1. Some trailers were simply ripped from official channels, with AI-generated thumbnails added on top. I’m not really considered these as “AI-generated.”

2. I mean, arguably we already have - I doubt the endless Facebook slop of Jesus made of various foods or African children showing off their implausibly elaborate crafts can be described as "an act of creative expression."

3. Though even this is half-hearted: both services allow you to remove the watermarks by upgrading your paid subscription.