SlopTV: Infinite AI Video Livestream Powered by Dual RTX 5090 GPUs
A developer has launched SlopTV, an interactive YouTube livestream that continuously generates and broadcasts 15-second AI video clips based on viewer chat inputs. The system runs fully locally using the open-weights MiniMax H3 video model powered by dual RTX 5090 GPUs. This project demonstrates the feasibility of running state-of-the-art open-weights video generation models locally on consumer-grade hardware. It also showcases creative, real-time interactive entertainment driven entirely by local AI workflows. The setup utilizes an int8 pruned diffusion model (19.5GB) and an nvfp4 text encoder (14.6GB), leveraging ComfyUI's VRAM offloading to manage memory. Clips are rendered at 352x608 resolution in about 90 seconds per GPU, alternating to output a new video every 45 seconds.
## BACKGROUND
MiniMax H3 (also known as Hailuo 3) is a multimodal AI model capable of generating high-quality video and audio from text or image prompts. ComfyUI is a popular node-based graphical user interface and backend for designing complex workflows with diffusion models. Running these large models locally often requires high-end GPUs and optimization techniques like model quantization (such as int8 or nvfp4) to fit within available VRAM.