RefMod's Likeness 'Breakthrough' Degrades MiniMax H3 Video Quality — Fix Inside
Kooky-Mode3047 · reddit · 2026-10-04
The author diagnosed why their MiniMax H3 videos suddenly became oversharpened and oversaturated: the new RefMod character-likeness tool slams reference images in as video frames, blowing past H3's documented input limits (≤9 images, ≤3 video/audio clips ≤15s, ≤12 mixed files), wrecking quality. Workaround: drop RefMod, use a 2x2 reference grid with native 1344x768, 15s, standard H3 sigma (12:3), seeds2 with the beta scheduler (.6, .6), and Spectrum to cut internal calls — 35 minutes per generation. Turbo LoRA users mask the degradation via heavily shifted sigmas.
More from Multimodal
- ComfyUI nodes add reference-image and phrase-level attention control to Qwen Image 2.1 — Capitan01R- · 2026-10-04
- Two ComfyUI Nodes Add Attention-Level Control to Qwen Image 2.1 Editing — feroszy · 2026-10-04
- Eleven v4 Delivers First Zero-Edit Voice Output in Blogger's Test, 9 Free Days Left — FellMentKE · 2026-10-04
- AdScanVideo Turns Any Uploaded Video Into a Timestamped, Structured Analysis Report — aftahi_ai · 2026-10-04
- A hands-on workflow for making Suno music videos with Claude — nptacek · 2026-10-04
- Higgsfield and OpenArt video generation tools are 'dangerously good' — MyCreativeOwls · 2026-10-04