A troubleshooting technique for AI video/image generation: when a prompt repeatedly causes the model to interact with the wrong object or element in a scene despite text description, draw a red arrow directly on the reference image pointing at the intended target. The visual annotation disambiguates spatial reference in a way text alone doesn't reliably resolve — e.g., a character pressing the wrong button on a prop sheet with several similar controls gets fixed once the correct button is marked.
Demonstrated in a Higgsfield/Hexel workflow (Cinema Studio, Seedance 2.0 4K) where an arrow was added to a prop reference sheet to fix a character interacting with the wrong element. Complements reference-prep techniques like Clean Your Plate (Reference Image Prep for Video Generation) and 3/4 Angle Establishing Framing — those improve overall reference quality, while this fixes a specific, recurring interaction-target failure.
Из тем: Holding Continuity Across Independently Generated Shots