Independent explainer · checked Sep 3, 2026
Infinite Slop: television that does not exist yet
It is not one infinitely long AI video. It is a public channel that plays one short scene while preparing the next — and lets the crowd compete to decide what comes after that.
Lightloom does not operate the stream, queue or chat. We explain the experience and link you to the original.
NOW the city folds into a paper ocean
- NEXTa moonlit train enters the oceangenerating
- Q 02the conductor is secretly a catwaiting
- Q 03turn the dining car into zero gravitywaiting
The useful definition
A television channel with the audience in the writers’ room.
The product is the shared waiting, voting, surprise and reaction — not simply the ability to press Generate.
- What is it?
- One shared, endless AI TV channel
- Who made it?
- Pieter Levels, with fal sponsoring inference
- Can I steer it?
- Yes — suggestions compete for public airtime
- Where do I watch?
- InfiniteSlop.ai is the official channel
Primary references: official channel · creator’s launch account
How the channel stays alive
Four states. One loop. No final episode.
- 01
CHAT
Suggest
A viewer describes what should happen next. - 02
QUEUE
Compete
Ideas wait for a limited slot. Submission is not a guarantee. - 03
H3 MAX
Generate
A selected idea becomes a short clip with synchronized sound. - 04
LIVE
Air
Everyone sees the same result while another clip is prepared.
*For a successful run near fal’s published 15-second timing. Network, queueing, failures and prompt processing can still add delay.
The product shift
The scarce thing is airtime, not the Generate button.
One request. One file.
- Audienceone person
- Resulta private asset
- Waitingfriction to remove
- Finish linedownload
Many requests. One show.
- Audiencea crowd
- Resulta public moment
- Waitinganticipation
- Finish linethere isn’t one
At fifteen seconds per scene
4possible fresh slots per minute
before retries, replays or transition time
Do not mix these up
Model, engine and show are three different things.
Calling all three “H3” hides where the experience actually comes from. The model makes pixels; the product makes television.
- 01MiniMax H3
- also branded Hailuo H3
The open-weight base model family.
- 02fal H3 Max
- the fast generation engine
fal's hosted, further-trained H3 variant for short clips.
- 03Infinite Slop
- the audience-facing product
Queue, selection, continuity, playback and a crowd sharing one channel.
Need the speed, quality and feature limits instead? Read the plain-language fal H3 Max guide →
Retail-equivalent cost
“Forever” gets expensive before the first day ends.
fal lists standard H3 Max at $0.05 per output second for 480P and $0.08 for 768P. Apply that rate to a day of entirely fresh footage:
480P: about $4,320/day
768P × 30 days: about $207,360
These are illustrations from public output rates, not Infinite Slop’s actual bill. The creator says fal sponsors inference. Delivery, storage, moderation and failures are outside this math.
Rate and timing source: fal’s current H3 Max guide
Why trust this explanation
Every claim has a status, not just a confident tone.
We checked the live product, the creator’s account and the model provider separately. Unknowns stay unknown.
The product
The official channel exposes chat, queue, generating and on-air states.
The creator account
Pieter Levels says viewer text becomes a later scene and fal sponsors inference.
The speed
fal reports five seconds of 768P output can finish inference in under three seconds.
The hidden parts
Exact selection logic, moderation stack and the project's actual sponsor cost are not disclosed.
Checked September 3, 2026 · See our editorial and corrections policy →
Try the smallest useful version
Make one scene before you imagine an endless channel.
Lightloom currently runs standard MiniMax H3 — not H3 Max. Write the scene here, see the exact credit cost before submitting, then sign in only when you are ready to render and keep it in your history.