MiniMax H3 Low-VRAM Video

A practical agent skill for deploying, running, troubleshooting, and improving MiniMax H3 local video generation on Windows PCs with limited NVIDIA VRAM and system RAM.

Use it for text-to-video, image-to-video, first/last-frame generation, reference-conditioned video, native audio, and preparing MP4 files for sharing.

Included:

• Hardware, storage, pagefile, and runtime checks before deployment.
• ComfyUI installation and model-selection guidance.
• Conservative low-memory startup settings and a staged resolution workflow.
• Prompting advice for clearer faces, stable motion, and coherent audio.
• Troubleshooting for model loading, out-of-memory errors, slow sampling, decoding, and MP4 muxing.
• A tiled, frame-by-frame Real-ESRGAN upscaling script that retains the original audio.
• Guidance for validating frame counts, codecs, dimensions, duration, and complete media decoding.

Requirements: a Windows PC with an NVIDIA GPU, a compatible ComfyUI/PyTorch setup, the appropriate H3 model files, and sufficient disk and virtual-memory capacity. Models and application runtimes are downloaded separately. Performance varies with hardware, settings, and software versions.

The skill distinguishes native video generation from an enhanced 1080p publishing copy; upscaling does not add native 1080p detail.