MiniMax H3 in ComfyUI: Generates 10s Video in 10 Mins on a 4090
wjc_5 · reddit · 2026-08-05
The author tested the newly open-sourced MiniMax H3 video model in ComfyUI, focusing on the full-reference workflow and local generation speed optimizations.
Performance Optimization
On a 48GB 4090, using the INT8 pruned model and NVFP4 text encoder to generate a 10-second video at 1 megapixel:
- Without acceleration: 30-34 minutes
- With SageAttention: 14-16 minutes
- With SageAttention + Prediction node: 10 minutes
Workflow & Tips
- The model supports up to 9 images, 3 videos, and 3 audio clips as reference inputs.
- The author noted that the Prediction node might cause visible softness on distant subjects. For high-motion scenes, it is recommended to use acceleration to find a good seed, then disable prediction for the final run.
- A system prompt template for web-based vision LLMs is provided to help organize rough stories and reference materials into precise H3-ready prompts.
Related event: MiniMax H3 Local Tests: Performance and Optimization Across GPUs(13 posts)→
More from Multimodal
- Fish Audio's open-weight Drama 3 voice model claims top emotive quality across 83 languages — mhdfaran · 2026-10-16
- Claude Opus 5.5 makes a pure-code Halloween animation starring bots trick-or-treating for tokens — sattvi · 2026-10-16
- 1,290-line commit lands MOSS Image support in ostris/ai-toolkit — ostrisai · 2026-10-16
- Ostris AI Toolkit adds training support for OpenMOSS's MOSS-Image-SFT model — ostrisai · 2026-10-16
- Nokia vs iPhone, car edition: AI image shows Tesla Roadster hovering over a Porsche — vaibhavbetter · 2026-10-16
- Midjourney multi-sref weight mixing recipe for cinematic light & motion portraits — michaelrabone · 2026-10-16