Global

FreeVideo is a local inference engine for MiniMax H3 on consumer GPUs, built on OpenVDN's 8-step VDN-H3 model with Video DeltaNet's hybrid attention.

5: Refresh RefMods updates the current Loader/Axis dropdowns from the RefMod folders without refreshing all ComfyUI models. Selections and connected inputs are preserved. The library's Refresh also updates that loader's dropdowns.

> Breakthrough inference speed for MiniMax H3 Reference-to-Video (Ref2VA / R2VA): Generate a full 5-second, 24 fps video conditioned on reference character images in 72s (warm) / 95s (first run) on a single consumer NVIDIA GeForce RTX 4090 (24GB).

Start with the MiniMaxH3 All-in-One Full Timeline Director workflow: one lightweight graph covers text-to-video, image reference, audio-driven generation, video editing, character replacement, motion transfer, digital humans, and manually segmented long-video generation.

MiniMax H3 renders faces poorly when the head occupies a small fraction of the frame. This is a property of head-size-in-frame, not of output resolution, so it persists at 720p and above.

This project has not provided a description yet.

WARNING] > This project targets an undocumented / pre-release Neural Rendering interface and is not affiliated with, endorsed by, or supported by NVIDIA or Comfy Org. dll versions. Native driver/runtime failures can crash the ComfyUI process.

4B-parameter image super-resolution model (LightningDiT + Qwen-Image 2D VAE + DINOv2-L conditioning). This is a standalone community package; it is not part of the upstream cswry/VOSR repository.

Running ONNX/TRT version of the MiniMax-H3 VAE in ComfyUI, which increase speed by up to 1.7x

Modern PyTorch/CUDA can shuffle weights between RAM and VRAM on demand, so in principle you can generate without tiling even when the working set exceeds VRAM. In practice this is dramatically slower, and the reason is a bandwidth gap of one to two orders of magnitude:

> Disclaimer. video2dlssnr is not affiliated, associated, authorized, endorsed by, or in any way > officially connected with NVIDIA Corporation or any of its subsidiaries or affiliates.