FreeVideo launches MiniMax H3 on a home GPU

The open-source FreeVideo project by FlashML-org released v0.4.4 of its local engine and ComfyUI plugin, which generate video using the MiniMax H3 model on home hardware with as little as 8 GB of VRAM and 16 GB of RAM.

Compiled by Sergey KostenchukPublished 2026-10-14Updated 2026-10-14

2026-10-14 Coding Apple

💻 FreeVideo launches MiniMax H3 on a home GPU

The open-source FlashML-org project released v0.4.4: a local engine and ComfyUI plugin that generate video using the MiniMax H3 model on home hardware with as little as 8 GB of VRAM and 16 GB of RAM. Launchers are available for Windows, Linux, and macOS (Apple silicon, preview).

🌍 The barrier to entry for high-quality video generation is shifting from server-grade B200/H200 GPUs to consumer cards — a direct challenge to subscription-based cloud services. The code and weights are open, but MiniMax's Community License includes territorial restrictions.

👤 FreeVideo.exe automatically sets up ComfyUI, the environment, and models, and supports LoRA, first/last frame, and reference inputs. Realistically: 16 GB of VRAM — 8–10 minutes for a 10-second 768p clip; 32 GB — about two minutes.

Source 1: https://github.com/FlashML-org/FreeVideo Source 2: https://github.com/FlashML-org/FreeVideo/blob/main/docs/execution-planning.md

Sources