MiniMax-H3 I2V Test: ref2va Matches fl2va with Single Image Input

spacemidget75 · reddit · 2026-08-04

The author tested the ref2va mode of the MiniMax-H3 model and found that with just a single input image, its Image-to-Video (I2V) performance works just as well as using the fl2va mode, if not subjectively better.

Given the differences in model sizes, the author wonders if it's worth keeping the fl2va mode unless first-frame and last-frame specifications are strictly required, seeking further community validation.

Original post →

More from Multimodal

Multimodal channel →