The fal.ai Max Model Behind Infinite Slop

Infinite Slop is not just a cool interface. Under the hood, it relies on a video generation model fast enough to produce clips faster than they are watched. That model is called Max.

From Minimax H3 to Max

Max is a fine-tuned version of Minimax H3, a text-to-video model known for generating detailed motion from a prompt. By fine-tuning it specifically for the Infinite Slop use case, fal.ai created a variant optimized for speed, consistency, and the kind of open-ended prompts that come from live chat.

The result, according to the project's announcement on levels.io/infinite-slop, is roughly a 50x speedup. That is the difference between a prototype that stalls after every prompt and a real live stream that can play continuously.

Why video model speed matters for live streams

Video generation is expensive. Most models take seconds or minutes to produce a clip, which is fine for offline editing but unusable for live broadcasting. A viewer expects a stream to flow in real time, without long pauses between scenes.

Max solves that by compressing generation time to the point where the pipeline can stay ahead of playback. The system can queue the next clip while the current one is still playing, creating the illusion of a truly infinite broadcast.

Sponsorship and infrastructure

Running a 24/7 AI live stream is not cheap. Inference, hosting, and bandwidth all add up. That is why fal.ai sponsors Infinite Slop. They provide both the model infrastructure and the engineering effort that made Max possible.

The future of Max and generative broadcasts

Projects like Infinite Slop are early signals of a new media format: generated, interactive, always-on broadcasts. As models like Max become faster and cheaper, expect to see more channels, shows, and experiences that are created live by AI rather than filmed in advance.

Want to see Max in action? Open the Infinite Slop live stream and send a chat message.