Interactive keyboard-controlled video model now runs at 5B parameters on consumer GPUs

multimodalart · x · 2026-07-29

A demo of an interactive, real-time, keyboard-controlled video model that now comes in 5B parameters — not 0.5B — and still runs on consumer GPUs.

The author says the model was successfully shown on ZeroGPU Spaces and can also run locally, making it notable as a lightweight but usable real-time video generation/control system.

Related event: Real-Time Keyboard-Controlled Video Model Runs on Consumer GPUs(2 posts)→

Original post →

More from Multimodal

Multimodal channel →