Your reference frame.
Video that respects it.
Text to video invents a new subject on every render. Starting from a reference image anchors it, which is the difference between usable footage and a nice clip of a stranger.
The Anchor Problem
Ask a video model for the same character twice from text alone and you get two different people. That is fine for b-roll and useless for anything recurring.
A reference frame removes the guesswork. The model is not deciding what the subject looks like, it is deciding how the subject moves.
Consistency You Can Build On
Recurring characters, brand assets and product footage all need the subject to survive between renders. Reference-anchored generation is the only approach that gets you there reliably.
Describe Motion, Not Appearance
Because the frame handles appearance, your prompt is free to be about movement, camera and pacing. Prompts get shorter and results get more predictable.
Frequently Asked Questions
What Makes a Good Reference Frame?
Sharp, well lit, subject clearly separated from the background. Whatever is ambiguous in the still stays ambiguous in the video.
Can I Use a Generated Image as the Reference?
Yes, and that is a common workflow. Lock a character first, then animate the frame you liked.
How Much Motion Can I Ask For?
Moderate movement holds well. Large, fast action gives the model more room to break the subject.
Anchor your video to a frame you chose
Your first generation is free.