Infinite Slop adds viewer voting to its live AI-video queue
Infinite Slop now lets viewers upvote what its AI-video stream generates next and has passed 2,000 concurrent viewers. Its creator says the project would cost about $200,000 a month without FAL sponsorship.

TL;DR
- Viewer prompts now move through a public upvote queue, with the most-supported item sent to generation and then aired, as levelsio's 2,000-viewer update lays out.
- The experiment crossed 1,000 simultaneous viewers in levelsio's 1,000-viewer update, then drew 37,000 viewers over one day, according to levelsio's first-day update.
- fal's H3 Max made continuous programming plausible: levelsio's launch post says the tuned model can make 15 seconds of video in nine seconds, although the channel caps itself at four 15-second clips per minute.
- The stream's reported running cost is about $4,000 a day, according to levelsio's daily-cost reply.
fal's H3 Max browser tool says it can turn a prompt or image into a 768p clip in as little as three seconds. A single banana prompt can spread through the channel's output, levelsio said, and levelsio's archive reply says every segment is kept.
The voted queue
Chat is now a programming surface rather than a comment rail. The live loop has four stages:
- A viewer types a prompt.
- The prompt enters the queue for other viewers to upvote.
- The most-upvoted entry is pulled into generation.
- The completed clip is placed next in the stream.
The selection rule changed during the rollout. The initial queue selected the oldest submission, levelsio's FIFO reply said, before voting became the published mechanism.
The portrait player
The player moved from landscape to 9:16 after traffic showed a heavily mobile audience.
levelsio asked whether mobile visitors should land directly in chat or first see the video behind a Chat button in levelsio's mobile UX poll. He later said he would retain the existing arrangement in levelsio's follow-up.
The live UI separates PLAYING from NOW GENERATING.
A later patch showed each viewer where their request sat in the queue, levelsio's queue update said. Older clips may become themed sidebars, levelsio's sidebar reply suggested, while levelsio's channels reply agreed with categorizing them into left-side channels.
The build remains visibly in motion: levelsio's reply marked a viewer suggestion for addition, levelsio's fix reply confirmed a reported problem was fixed, and levelsio's prompt question asked a viewer what they had submitted. levelsio described watching the channel as "weirdly addictive" in a reply.
Four independent clips
The launch pitch said each new chat request would try to connect to the preceding video, according to levelsio's original announcement. The underlying model produces video and sound together, levelsio's audio reply said, while fal describes H3 Max as a post-trained MiniMax H3 variant optimized for prompt adherence, aesthetics, and throughput in its API documentation.
The system currently keeps four clips independent and generates them in parallel. In levelsio on frame chaining, the creator gave five reasons that using every clip's final frame as the next clip's starting frame would break the live format:
- It serializes four workers into one dependency chain.
- The next chat prompt does not exist at the live edge.
- A fresh viewer request inherits an unrelated scene.
- Voting, VIP jumps, and newest-first ordering can no longer reorder clips.
- One failed generation stalls all subsequent clips.
A quiet, deliberate story continuation is the exception because it can run one clip at a time. In his speed note, levelsio put H3 Max at 15 seconds rendered in nine seconds, but seamless frame chaining would trade that throughput for continuity.
The runtime bill
fal's H3 Max API documentation lists a promotional 768p rate of $0.04 per generated second through September 1, then $0.08 per second. At continuous output, that is $144 an hour or $3,456 a day before the change, then $288 an hour or $6,912 a day afterward.
Those numbers bracket levelsio's $4,000-a-day estimate and his statement that the project would cost about $200,000 a month without fal sponsorship, in levelsio's sponsorship-cost reply. He put the profitability threshold at $210,000 a month.
levelsio rejected a proposal to let viewers bid for queue position, saying that would be less fun than the current free-for-all in his first-day update.