MiniMax H3 refmod labels inconsistently switch between Picture and Video references
Azhram · reddit · 2026-10-28
A user experimenting with refmod (via a Civitai MiniMax H3 refmod-creator workflow) hit inconsistent labeling: the first 6-image reference set was correctly tagged <Picture 1>, but later sets got referenced as Video x — even when using 6 screenshots. The pattern seems to depend on image content or ordering, and the user can't figure out how the model decides, making switching between sets inconvenient.
More from Multimodal
- Google exec teases Genie 4, its new frontier world model, coming to the API — OfficialLoganK · 2026-10-28
- Voxelizing OPLONTIS: AI reconstructs ruins one-to-one with measured voxel cubes — Promptmethus · 2026-10-28
- VybBench: open benchmark scores image models' design fidelity with numbers, not votes — thetripathi58 · 2026-10-28
- Midjourney multi-sref blending recipe shared, with full prompts — creatoroff · 2026-10-28
- Leak: DeepMind's Genie 4 is a real-time world model with native audio and type-to-edit scenes — rohanpaul_ai · 2026-10-28
- The "Muse Cinematic Universe" grows: AI-generated short film goes viral — inductionheads · 2026-10-28