Infinite Slop
An endless, audience-directed AI television channel
Viewers submit ideas, the crowd decides what reaches the queue, and AI turns the winning prompt into the next shared scene. It is closer to interactive television than a private video generator.
Information checked August 31, 2026
Watch Infinite Slop live
The live service is embedded directly below. Playback, chat, verification, moderation and availability are operated by Infinite Slop inside the frame.
Loading the live channel…
What is Infinite Slop?
Infinite Slop is a continuous AI-generated channel created by Pieter Levels with inference sponsored by fal. The audience does not receive a private download: prompts compete for public airtime and the selected scene becomes part of a stream everyone watches together.
The scarce resource is airtime, not the Generate button
When thousands of viewers share one channel, only a handful of prompts can air each minute. Queue ranking, voting and anticipation become the product—not just model output.
| Typical AI video generator | Infinite Slop |
|---|---|
| One user waits for a private file | A crowd watches one public stream |
| A request ends after download | Every result becomes the next live event |
| No competition for output | Prompts compete for queue position and votes |
| Generation latency interrupts the task | A playback buffer keeps the channel moving |
| A creation tool | A participatory media product |
The channel turns prompting into a multiplayer loop
The interface combines familiar community mechanics with faster-than-playback video generation.
Chat becomes programming
A short message is not only commentary; it can propose the next event, character or genre for the channel.
The queue creates scarcity
Now Generating, Playing Next and On Air make waiting visible and turn limited broadcast slots into a shared game.
Votes choose attention
Upvotes allow the audience to rank ideas before the system spends inference time on them.
Categories create channel culture
Banana, cats, space, news and game-show filters turn repeated motifs into recognizable communal formats.
Continuity creates feedback
The system tries to connect a new scene to the previous one, so visual motifs can persist and influence later prompts.
Mobile framing changes the medium
A vertical feed makes the result feel less like a desktop livestream and more like an always-running generative short-video channel.
How a viewer idea reaches the screen
The exact private backend is not public, but the observable product flow is clear.
Viewer prompt Human check and moderation Public queue and votes Selected prompt Context and continuity H3 Max generation Playback buffer Shared live stream Audience reaction
The last step feeds the first: viewers react to what the model produced, then write the next prompt. That closed loop is the difference between generated content and generative media.
H3 Max makes faster-than-playback video practical
H3 Max is fal's post-trained MiniMax H3 variant, co-optimized with its inference stack. Infinite Slop demonstrates what changes when a model can prepare the next clip before the current one finishes playing.
Faster than real time
fal reports that a five-second 768p video can render in under three seconds on its optimized stack.
Video and sound together
The model keeps MiniMax H3's native synchronized audio-video capability, so the stream does not need a separate sound-generation pass.
Text and image modes
H3 Max launched with text-to-video and image-to-video endpoints, including an optional ending image for visual transitions.
Up to fifteen seconds
A single generation can run to fifteen seconds, giving the channel a natural scene-sized broadcast unit.
The economics are closer to broadcasting than personal generation
One generated clip can serve one viewer or one hundred thousand viewers. Inference is a large fixed content cost; audience growth mainly adds delivery, chat, storage and moderation costs.
24/7 retail API cost scenario
Estimate the retail-equivalent cost of continuously generating one second of 768p output per second. This is not Infinite Slop's reported spend: fal sponsors the project, and private commercial terms may differ.
Pricing snapshot checked August 31, 2026. The launch discount was scheduled to end September 1. Bandwidth, storage, moderation and failures are excluded.
A product category appeared in a weekend
The sequence matters because it shows how rapidly a new model capability can become a distinct media format.
fal releases H3 Max
The post-trained model crosses the faster-than-playback threshold in fal's published benchmark.
Infinite Slop goes live
Pieter Levels publishes an interactive stream inspired by earlier community experiments and says fal sponsors inference.
The dedicated domain follows validation
After reporting about 37,000 viewers on the first day, the creator registers InfiniteSlop.ai.
Queue and voting become the core interface
The live site presents categories, chat, queue states and audience voting around scarce broadcast time.
Why this experiment matters beyond one website
The durable idea is not endless random clips. It is a live world that viewers can collectively steer while it is being rendered.
Generative television
A channel can continue indefinitely and adapt its subject, characters and tone to live audience input.
Prompt battle formats
Rounds, voting and limited airtime can turn prompting into a spectator competition without running a full 24/7 stream.
Persistent vertical channels
Anime, horror, children's stories, game shows or branded worlds can develop their own recurring rules and community memes.
Interactive worlds
Longer memory and stable characters could move the format from disposable clips toward collective role-playing and ongoing narratives.
What is not solved yet
Faster inference removes the largest wait, but a continuous public medium creates a new class of operational problems.
Continuity versus buffering
Waiting for the exact final frame before generating the next clip improves continuity but risks dead air when any request is late.
Moderation at broadcast speed
Prompts and outputs need checks before thousands of people see them, without turning every transition into a long pause.
Queue fairness
With only a few slots per minute, voting can reward popularity, repetition or organized groups instead of creative variety.
Retail inference cost
A true always-on channel can consume six figures of monthly API value at public rates, before delivery and operations.
Third-party availability
This embedded channel can change, require verification or go offline independently of Wan2.Video.
No public official source code
The current Levels version is not known to have an official open-source repository, so backend implementation details remain inferred.
Infinite Slop questions answered
Clear answers on the product, model, access, cost and source code.
What is Infinite Slop?
It is an endless AI-generated television channel where viewers propose and vote on prompts that can become the next public scene.
Is Infinite Slop a normal AI video generator?
No. A normal generator returns a private asset. Infinite Slop turns a limited number of winning prompts into one shared continuous broadcast.
Which model powers Infinite Slop?
The creator identifies fal's H3 Max, a post-trained and inference-optimized MiniMax H3 variant with synchronized audio and video.
Is Infinite Slop free to watch?
The embedded public channel is currently viewable without a Wan2.Video account. Chat, voting, verification and availability are controlled by the third-party service.
How can it generate continuously?
H3 Max can render a five-second clip in under three seconds in fal's published benchmark. Generating ahead and buffering clips lets playback continue while the next scene is prepared.
How much could a 24/7 stream cost?
At US$0.04 per output second, one uninterrupted stream equals about US$3,456 per day; at US$0.08 it equals about US$6,912. This is a retail-price illustration, not the project's actual sponsored spend.
Is Infinite Slop open source?
No official public repository for Pieter Levels' current implementation has been identified. Other generative livestream projects may be open source, but they are separate products.
Who created Infinite Slop?
Pieter Levels built the current website and credited Marc-Antoine and Rehan Sheikh's earlier stream experiments as inspiration. fal sponsors inference.