Single-GPU MiniMax H3 live video workflow cuts VRAM from 80GB to under 30GB
Overview
A ComfyUI developer, Bryan Wade, describes a workflow that generates 15-second 448×256 video in 15 seconds or less on one RTX 5090 using MiniMax H3 with FastVideo's FastH3 V2 checkpoint at four sampling steps.
The developer says the setup combines sparse attention, a smaller ClipProj text encoder, a pruned INT8 checkpoint, and a fused FP4 MLP, reducing VRAM needs from 80GB to under 30GB. The custom ComfyUI node is open source.
These figures come from the developer's first-party blog post and have not been independently verified. The post does not report output quality comparisons against the full-size model.
Written by AI from the articles below · updated Oct 8, 7:52 PM ET
Check the sources:
Article timeline
Follow the coverage from different perspectives. Times are ET.
- Comfy BlogHow I Generated Live Video with MiniMax H3 on a Single GPU
AIA ComfyUI developer generated 15-second 448×256 video in 15 seconds or less on one RTX 5090 using MiniMax H3 with FastVideo's FastH3 V2 checkpoint in four sampling steps. The setup combined sparse attention, a smaller ClipProj text encoder, a pruned INT8 checkpoint, and a fused FP4 MLP, cutting VRAM needs from 80GB to under 30GB. The custom ComfyUI node is open source.
Heat trend
Not enough continuous observations to show a trend yet.