The purpose of these merges is surface rendering: surfaces, textures, fine details and foliage. Z-Image's spatial attention profile - how sharply it commits to fine texture - is transplanted onto video engines by rescaling their attention normalisation. Identity, motion, sound and speed stay untouched. The difference is paused-frame visible: straw holds strands at mid-distance, hedges render as brambles and leaves, fabric keeps its weave, skin keeps its grain.
All comparisons are same seed, same settings, matched timestamps. Generated content is machine-generated. Community licenses of the respective base models apply.