Which reference inputs can AI video models use?
Updated August 2026
Ask an AI about this guide
Open this article in your assistant with a ready-made question.
Reference inputs give a video model more structure than text alone, but no single model accepts every input type. Use only the controls exposed for your selected model and route.
Frames define composition and continuity
A first frame anchors how the clip begins. A supported last frame also gives the model a destination, which is useful for transitions and controlled changes over time.
References and source clips serve different jobs
Reference images, video or audio can guide appearance and behavior where supported. A source-video input is for model-supported editing or extension, not a universal control available on every route.