Spider-Verse 2025: Streaming 24/7. No Reruns. Forever
The future of AI-powered entertainment is more of everything. A lot more. Infinity more.
If you don’t have small kids or you aren’t around people who have small kids, you may not have seen the future of entertainment, yet. I saw it the other night at a Halloween party, where it was playing on a screen in another room. Here it is, take a look:
This kind of content is all over YouTube, and going by the astronomical view counts these vids rack up (20.7M views for the first, and 16.1M views for the second) it’s quite popular.
Obviously, these low-effort videos are just total garbage. In fact, they’re so awful and disturbing that I personally consider exposing your kids to them a form of child abuse. But as junky as these videos are, they offer an important glimpse at the future of online entertainment in the era of cheap, ubiquitous AI content generation.
Imagine a world where Disney has fully embraced generative machine learning models internally across all its content units. The company has immense resources to throw at the project of training a suite of customized models — text and story generation, text-to-image, text-to-3D, text-to-video — on its vast library of existing content.
Let’s narrow our scenario down to one property: imagine that Disney and Sony have collaborated to produce a family of Spider-Verse models trained on every piece of Spider-Man content since the character’s first appearance in a 1962 issue of Amazing Fantasy #15. Imagine that in the year 2025, all of these models are stitched together by a very small team of humans (less than ten) to produce a Disney+ streaming channel called Spider-Verse 2025.
The #CurrentSpiderVerse of 2025
Welcome to the year 2025, where every comic, radio play, movie, short story, novel, cartoon, and whatever has been dumped into a set of machine learning models more advanced and powerful than anything currently available today. It’s a world where something like the following process goes on 24/7 in a Sony data center, pausing only for the occasional upgrade or bit of maintenance:
The Spider-Verse LLM generates a ream of scripts for next week’s streaming content. Each individual script is designed to support about an hour of video, and to form a discrete chapter in an ongoing, open-ended family of interlocking and (mostly) internally consistent narratives centered on different characters in the Spider-Verse.
The scripts are quickly vetted by a safety model that’s essentially an ML-powered sensitivity reader. The model can output structured feedback that the Spider-Verse LLM can take in and rework the script to produce something that’s marked “safe” for all audiences.
The vetted scripts are fed into a set of models that turn them into enough hours of audio and video to fill the entirety of next week’s 24/7 streaming Spider-Verse 2025 channel on Disney+ about three or four times over.
The resulting collection of chapters in the Spider-Verse’s ongoing narratives is fed into a set of rating models that are constantly being trained and retrained on engagement data from across Disney’s digital properties. These models can predict which chapters will perform the best with viewers in the channel’s target audience so that only the most engaging content makes it into one of the Spider-Verse 2025’s one-hour chapter slots.
Viewers can then tune into the Spider-Verse 2025 channel on Disney+ at any given moment of the day and see brand new, never-before-seen Spider-Man shows that are:
Internally coherent, with a recognizable narrative structure and some character development in each episode.
Consistent across episodes, so that a character that died in a chapter that aired last week stays dead, or a villain that changed teams and became a hero sticks to his hero arc for a certain number of stories.
Contain larger, multi-episode story arcs.
Shockingly good, as in, way better by any conceivable metric than the Saturday morning cartoon fare we Gen-Xers grew up with.
If you’re a Spider-Verse 2025 fan and you missed a few hours of streaming because you had to take care of some bodily function like sleeping or eating, you can always get caught up by reading a custom-generated summary of events that begins at whatever point in the story you last paused the stream and ends at the present moment. That summary could be pure text, or it could be a multi-panel comic book drawn in the style of Jack Kirby — it just depends on what you’re in the mood for.
Every year on National Spider-Man Day (August 1st), there’s a complete continuity reset as the channel name changes to the current year. So on August 1, 2026, Spider-Verse 2026 starts over with a story about a teen being bitten by a radioactive spider and features a new world with a new art style, new costumes, new characters, and new twists on old characters.
Of course, all the stuff that happened in Spider-Verse 2025 still happened in the continuity of the Marvel universe, but it just happened in a different Spider-Verse than the one you’re now watching in 2026.
Sometimes, the 2025 Spider-Man will show up in the 2026 stream, and at other times a Spider-Man from a future Spider-Verse will show up in the present stream. Sharp-eyed fans chronicle these crossovers and previews on the numerous blogs and social media outlets dedicated to the #CurrentSpiderVerse.
If you don’t believe this is possible, you’re wrong
I can already hear the objections being raised to the scenario above, and they all come from the same place, i.e., there’s no way an AI can generate plots that are internally consistent, characters that develop, or story arcs that span multiple episodes, or videos from scripts, etc.
If you think this way, you’re just wrong. And you’re wrong because you haven’t spent enough time with any of these tools to be surprised by them. Or, maybe you’ve spent enough time with these tools to run up against their limitations, and you imagine that those limitations are somehow inherent to machine learning or impossible to transcend.
All these objections are not only wrong, but it’s not even worth spending any real-time refuting. The ecosystem of tools themselves will provide all the counterexamples in due time. Whenever a new iteration of one of these tools comes out, there are always skeptics who find themselves scrambling for a new thing about which to say, “But it still can’t do that!” This goalpost migration is actually as reliable a part of the advance of AI as the advance, itself.
The other thing that makes Spider-Verse 2025 quite realistic in the near term is that the bar for this content is so low. It’s not just that the popular videos posted at the start of this article can already be trivially beaten in quality by GPT-3 output, but even the shows many of us grew up with were beneath the bar of what the very next generation of LLMs will be able to generate scripts for. I mean, here’s some GPT-3 output I just generated from the text prompt in italics:
Write a script for an episode of Spider-Man 2025.
INT. PETER PARKER’S APARTMENT - NIGHT
Peter Parker is lying in bed, awake. His eyes are wide open, but he’s not moving. He’s staring at the ceiling, lost in thought.
Suddenly, his phone starts ringing. He doesn’t move to answer it. After a few rings, it goes to voicemail.
A few seconds later, there’s a knock at the door. Peter still doesn’t move.
The door opens and Mary Jane Watson walks in.
MJ: Hey, Peter. I know you’re awake. I can see the light from under the door.
Peter doesn’t say anything.
MJ: Peter, what’s wrong? You’ve been like this for weeks. You’re not eating, you’re not sleeping, you’re not going to class.
Peter still doesn’t say anything.
MJ: Talk to me, Peter. Please. I’m worried about you.
Peter finally speaks.
Peter: I’m fine, MJ. Just leave me alone.
MJ: I can’t do that, Peter. I care about you.
Peter: I said I’m fine! Just leave me alone!
MJ: Peter, please.
Peter: I said leave me alone!
MJ: Fine. I’ll go. But I’m not giving up on you, Peter. I’ll be back.
MJ leaves and Peter is left alone again. He continues to stare at the ceiling, lost in thought.
Conclusion
I’ve centered this scenario around the Spider-Verse, but it’s easy to imagine content streams centered on any or all of the popular IPs that percolate in the endless digital zeitgeist of reboots, remixes, and trending hashtags. It’s also easy to imagine multiple streams from any given IP. Perhaps Spider-Verse 2025 is a family of 100 or more related streams, with different stories in different universes that sometimes cross over with one another.
Ultimately, something like what I’ve described here seems like the future once the larger IP holders fully embrace AI content generation. Alongside the discrete cultural products made (at least in part) by humans — blockbuster movies, prestige TV series, graphic novels, etc. — IP owners also run specially trained generative models to offer open-ended, always-on entertainment experiences set in their proprietary worlds, where fans can dip into an endless stream of engaging, AI-generated content that never dries up and never repeats.
You may wonder what becomes of creative professionals in such a world of infinite, AI-generated stories, but I think a more interesting and potentially more important question is, what becomes of fans and fandoms? What kind of culture can exist downstream of such content geysers that are constantly welling up from the deeps of our globe-spanning data centers?
How will people ever be able to get some serious work done in this future ? (just asking).
George Orwell‘s prescience continues to amaze.