MiniMax H3 / Hailuo 3.0 · Unknown

INT8 vs. W4A8 on Runtime, VRAM, and Hair-Sweeping Motion

With the same reference, prompt, and seed, INT8 and W4A8 are compared: W4A8 is about forty percent smaller on disk, but on an RTX 3060 Ti it runs slightly longer, shows no peak-VRAM reduction, and renders the hair-sweeping action less clearly. Output: 5s · 736×1280. The source did not publish a complete prompt, so this library does not infer or complete one.

Generation mode
Unknown
Model
MiniMax H3 (creator-labeled)
Output
5s · 736×1280 · portrait
Tags
Model Comparison · Comparative
Prompt provenance
Not published by the source

Prompt not published by the source
This page only shows the video and verifiable public information. It does not infer or complete a prompt from the video.

View original source · Original X post · @aihonobono2023