MiniMax H3 official samples look stunning, but ComfyUI reproductions fall far short

cptdurr · reddit · 2026-11-16

A Reddit user asks why MiniMax H3's 50+ official GitHub sample images look spectacular while ComfyUI reproductions with the same prompts differ drastically — worse fur understanding, lion facial structure, and more. They suspect sampler differences (MiniMax may not use Euler/Simpler defaults internally). The thread highlights the recurring gap between official showcases and community reproductions for image models.

Original post →

More from Multimodal

Multimodal channel →