Three-keyframe reference triptych
Split a video prompt into three complete standalone keyframes: start state, action midpoint, and consequence hold, then give Seedance references that already look like final frame… Inspect the generated proof and apply the visual system to a private Vixel Canvas Work.
Technique
Split a video prompt into three complete standalone keyframes: start state, action midpoint, and consequence hold, then give Seedance references that already look like final frames.
Why it is useful
This avoids asking GPT Image 2 for a text-heavy storyboard when the video model really needs coherent still references.
Technique source: @IamEmily2050.