Why did @levelsio build Infinite Slop?

What is Infinite Slop, exactly?

Infinite Slop is currently an endless AI-generated TV channel where viewers type ideas into a chat and influence what appears on screen next.

The stream keeps generating short AI videos one after another. Levels says the system also tries to connect each new generation with the previous one, so a prompt does not always produce a completely isolated clip. Characters, objects and visual ideas can spill into whatever comes next.

The live page describes the concept very simply: “an endless AI-generated TV channel” where “the chat decides what airs next.” In practice, it feels somewhere between Twitch, an AI video generator and a collective improv show. People suggest scenes, the system chooses what to generate, and everybody watches the result together.

That audience participation is the interesting part. Endless AI content existed before Infinite Slop. What is new here is how quickly viewers can push the stream in a completely different direction while it is already running.

Why did Levels build Infinite Slop now?

Levels built Infinite Slop now because Fal's H3 Max made continuous AI video fast enough to work as a live experience.

Fal says a five-second 768p H3 Max video can currently render in under three seconds. Its own benchmark puts throughput at roughly 35 times the official MiniMax H3 endpoint and about 15 times faster than comparable models. Levels used a different comparison in his post, saying the new version was around 50 times faster and showing 15 seconds of video generated in roughly nine seconds.

Those numbers are different enough that we should keep them separate. Fal's current documentation also says longer generations take longer, with a 15-second clip around the real-time mark in its standard setup. But the broader change is clear: at least some good-quality AI video can now be produced before the viewer has finished watching the previous clip.

That immediately creates use cases that were awkward before. A model taking several minutes to produce 15 seconds of footage works for making clips. Once generation approaches or beats playback speed, you can start thinking about channels that never stop.

Levels spotted that change almost immediately.

What happened	What it changed
Fal released H3 Max	Good-quality AI video became much faster to generate
Rehan Sheikh connected H3 Max to a livestream	Endless generated television became an actual working demo
Levels saw the experiment	He publicly highlighted perpetual livestreams as a new use case
Levels launched Infinite Slop	The audience could now steer the next scenes through chat
Levels kept modifying it live	Queueing and other fixes appeared within hours

Did Levels invent Infinite Slop?

No. Levels openly credits Marc Antoine for the idea and Rehan Sheikh for the original livestream that inspired Infinite Slop.

Rehan Sheikh, who works around Fal's video stack, had already connected H3 Max to Twitch and described the result as “infinite interdimensional cable.” That experiment demonstrated the core trick: keep generating footage quickly enough and a channel can theoretically continue forever.

Levels saw it and moved fast. His launch post specifically names both Marc Antoine and Rehan instead of presenting Infinite Slop as something he invented from scratch.

That tells us a lot about why the project exists. There was no long product-development cycle behind Infinite Slop. A technical capability appeared, somebody showed a rough version of what it could do, and Levels decided to build his own interpretation while the idea was still new.

This is much closer to opportunistic building than classic startup planning.

What did Levels actually add to Infinite Slop?

Levels's main addition to Infinite Slop was turning a perpetual AI stream into something the audience could actively mess with.

Anyone in the chat can suggest what should happen next. The system then takes some of those requests and generates upcoming footage while trying to preserve a loose connection with what was already playing.

That second part produces some surprisingly weird behavior. Soon after launch, Levels noticed that viewers kept asking for bananas. Because successive clips were being bridged together, bananas started appearing across otherwise unrelated prompts. A joke created by viewers effectively infected the stream.

That is more interesting than simply pressing “generate” repeatedly. The viewers start creating running jokes, recurring objects and accidental stories together.

Levels has built community-driven products before, especially Hoodmaps, where users fill a shared map with their own labels and opinions. Infinite Slop applies a similar instinct to generated video: give people a simple system, let them interfere with it, and see what culture appears around the mess.

Was Infinite Slop meant to become a real startup?

Infinite Slop currently looks much more like a fast experiment than a startup that began with a serious business model.

The public product is free. There is no subscription or normal payment flow visible in the current experience. Fal is covering the generation costs, and Levels launched the project through his experimental “skunkworks” setup rather than with the sort of polished commercial structure we would expect around an established SaaS product.

His behavior after launch points in the same direction. Within hours, he was still fixing issues, changing the interface and responding to people asking how the queue worked. At one point he reported roughly 400 people watching simultaneously. His own archived launch post has now passed 120,000 page views, so the experiment clearly found attention very quickly.

People around the launch have already proposed plenty of monetization ideas, including letting the highest bidder control the next scene, branded livestreams and generated advertising. None of those ideas currently define the product.

For now, Levels seems to be asking whether people find this format compelling before deciding what sort of business, if any, should sit on top of it.

Why does the chat matter so much in Infinite Slop?

The chat gives Infinite Slop its real hook because viewers can affect the television channel while they are watching it.

Passive infinite AI video has an obvious weakness: after the novelty wears off, viewers are left watching a machine generate random footage. Infinite Slop gives people something to do. You can try to derail the story, make everyone watch your absurd idea, continue somebody else's joke or wait to see whether the system understood what you typed.

That creates a completely different feedback loop from Netflix, TikTok or YouTube. Traditional platforms predict which existing video you are most likely to watch next. Infinite Slop can generate a video that did not exist when you opened the page.

There is also a multiplayer element. You are competing with other viewers for a scarce place in the stream, and everybody watches the winner's prompt at the same time.

That may matter more than the quality of the videos themselves. The early clips are inconsistent and often ridiculous. Yet the unpredictability becomes part of why people keep watching.

Can Infinite Slop actually keep up when lots of people join?

Infinite Slop can currently keep the broadcast moving, but viewer demand already exceeds the number of prompts the stream can show.

Levels explained that the first version could select four prompts per minute because it was producing four 15-second segments. Once more than four people want to generate something during that minute, most requests cannot immediately appear on screen.

That bottleneck showed up almost straight away. Within hours of launch, Levels added a queue system so viewers could see where their request was waiting. It is a small product change, but a useful one: the first scaling problem Infinite Slop encountered came from too many human requests for the available airtime rather than from a lack of things for the AI to generate.

The stream also has very little room for a slow generation. If one clip needs longer than expected, the system either needs a buffer or risks interrupting the flow. Levels discussed feeding the ending frame of one generation into the next to make transitions smoother, but that adds another dependency to a pipeline that already has to keep moving continuously.

Fal is now experimenting with a more ambitious solution. Its H3 Max Live work uses an experimental model that keeps context across scenes and preserves both audio and visual continuity. Fal says that version is coming to its API. If it works reliably, some of the continuity hacks Levels currently needs could disappear.

Who is paying for Infinite Slop, and how expensive is it?

Infinite Slop would cost well into six figures per month at Fal's standard public rates if it continuously generated 24 hours of video every day.

Fal currently lists promotional H3 Max prices of $0.025 per generated second at 480p and $0.04 at 768p. It says the normal rates are twice as high: $0.05 and $0.08 respectively.

Those numbers become large very quickly because an endless channel needs 86,400 seconds of fresh footage every day. At the standard 768p rate, the public-price equivalent comes to almost $7,000 per day. Thirty days would be about $207,000.

Levels is not paying that retail bill. He explicitly says Fal sponsors Infinite Slop, and Fal's internal compute cost will obviously differ from what an outside API customer pays. Still, the calculation explains why sponsorship is such an important part of the experiment. Running this continuously would be hard to justify as a free hobby at normal API prices.

H3 Max rate	Approx. cost per day of continuous video	Approx. cost over 30 days
480p promotional rate	$2,160	$64,800
768p promotional rate	$3,456	$103,680
480p standard rate	$4,320	$129,600
768p standard rate	$6,912	$207,360

Is Infinite Slop basically an ad for Fal?

Infinite Slop doubles as an excellent Fal demo, but we found no evidence that Fal commissioned Levels to build it.

The relationship between the two is much closer than a normal sponsorship. Levels lists FAL.ai in his personal angel portfolio, with his investment dating back to its Series A. He also uses Fal heavily inside Photo AI.

Levels previously explained that every AI model used by Photo AI runs through Fal's infrastructure. At the time, Photo AI was producing more than one million images per month plus some videos, which meant Fal was already an important supplier to one of his biggest businesses.

Infinite Slop now showcases the exact thing Levels has praised Fal for before: speed. Viewers do not need to understand inference benchmarks to see the point. They can watch generated footage appear continuously in front of them.

Fal has gone even further since Levels launched the project. The company is now showing its own H3 Max Live stream where chat controls an infinite broadcast, and Fal engineers are testing a version with native continuity between scenes.

So the incentives line up extremely well. Levels gets free compute for an experiment he finds interesting. Fal gets a very visible demonstration of its newest technology. Levels also owns a stake in Fal.

We do not have evidence to call Infinite Slop a disguised paid promotion. It is obviously commercially useful to Fal.

Why call it Infinite Slop if Levels has criticized AI slop?

The contradiction is real, although Infinite Slop seems to be deliberately leaning into the exact thing Levels has been criticizing.

Levels has become increasingly negative about the flood of interchangeable AI software and content. After reviewing more than 80 startup applications recently, he complained that everyone could now create decent-looking products while many of them felt almost identical. He has also argued that there are more apps but fewer sales, and lately he has warned that AI-generated content could push the internet closer to the old “Dead Internet” idea.

Then he launches something literally called Infinite Slop.

The name makes the joke impossible to miss. Levels does not pretend the stream is premium cinema or meaningful human art. He describes the output as slop himself and has called the whole experience dystopian.

What seems to interest him is what happens once the quantity and speed become extreme. AI slop normally means a flood of cheap posts, images or videos uploaded after generation. Infinite Slop turns that flood into something live, shared and steerable.

We can disagree about whether that produces better culture, but technically it is a different experiment from generating another generic AI app or filling social feeds with automated clips.

Does Infinite Slop fit how Levels normally builds products?

Yes. Infinite Slop fits Levels's actual project history much better than you would guess from looking only at Nomads, Remote OK or Photo AI.

Levels maintains a public database of everything he has built. In the latest version before Infinite Slop launched, he counted only nine projects as long-term financial successes. Another 11 made money or took off without becoming durable successes, while 19 failed.

The biggest category by far contains 69 projects that he labels N/A because they were non-profit or had no profit goal. That includes all sorts of experiments, art projects, games, utilities and strange internet ideas.

That context makes Infinite Slop look much less unusual. Levels does not spend all his time building businesses. He spends a lot of time making things, and only a small minority eventually become serious businesses.

Recently he has revived old Quake maps, put multiplayer retro games in browsers, built small AI utilities and continued changing Hoodmaps. Infinite Slop belongs naturally in that stream of projects.

The business strategy comes from the volume. If something unexpectedly catches on, he can keep pushing it. If nobody cares, another experiment replaces it.

How Levels classifies his projects	Number of projects
Long-term financial success	9
Took off but was not sustainable	11
Failed	19
No profit goal / N/A	69
Too new to judge	6

Is Infinite Slop actually a new kind of AI television?

Infinite Slop is a real step toward interactive generative television, even though people were already experimenting with endless AI shows years ago.

Nothing, Forever is the obvious earlier example. The AI-generated Seinfeld-style Twitch show became a viral hit and reached around 15,000 concurrent viewers. Its dialogue and scene direction were generated with AI, while the visual environment came from a reusable Unity-based 3D world.

Infinite Slop can change the actual moving imagery every few seconds. Ask for a completely different place, character or situation and the visual world itself can be regenerated. The show no longer needs a fixed apartment, character model or library of scenes.

The fresh development from Fal pushes the idea further. Its experimental H3 Max Live system is now trying to retain context across an indefinite stream instead of repeatedly restarting generation from short clips. Fal engineers describe it as native infinite continuity for both visuals and audio.

Put those pieces together and it starts to look much closer to genuinely generated television: live imagery, persistent context, viewer direction and no predetermined ending.

Whether anyone will want to watch that for hours is still wide open. Nothing, Forever showed that absurdity alone can attract thousands of viewers, but it also showed how quickly novelty and moderation problems can become part of the product.

So why did @levelsio build Infinite Slop?

Levels built Infinite Slop mainly because he saw AI video cross a technical threshold and wanted to find out what became possible once generation was fast enough for a live audience.

The chain of events is unusually easy to reconstruct. Fal released a much faster version of MiniMax H3. Rehan Sheikh immediately turned that capability into an endless Twitch stream. Marc Antoine contributed the idea that Levels credits. Levels saw the experiment and quickly built a version where viewers could steer the output themselves.

Everything that happened afterward looks like live exploration. Hundreds of people arrived at once. Bananas began contaminating unrelated scenes. Too many prompts piled up, so Levels added a queue. People immediately started proposing bidding systems and advertising models. Fal then pushed the same idea further with its own continuous H3 Max experiment.

There are commercial benefits around the edges. Fal is a company Levels invested in, Fal already powers Photo AI, Fal pays the compute bill here, and Infinite Slop gives the company a memorable product demo. Levels also has a huge distribution advantage, so he can expose a weird prototype to a large audience almost instantly.

Those incentives help explain why this was particularly easy for him to try, but they do not look like the original reason for building it.

Levels has dozens of projects that were never intended to become businesses. Infinite Slop fits that pattern extremely well. He saw something technically new, built the strange consumer version almost immediately, and started watching what people did with it.

So the direct answer is that @levelsio built Infinite Slop because faster AI video suddenly opened a new interactive format worth testing. The project may eventually become a business, an ad format, a creator platform or nothing durable at all. Right now, the experiment itself appears to be the point.