I was testing a six-second product clip and kept making the same mistake: adding too many instructions. The brief included a scene change, subject motion, a camera push-in, a lighting shift, and a final reveal. Each request made sense on its own, but together they made the result feel hurried.
While testing the sequence in the independent web tool MiniMax H3, I got a cleaner result after treating the clip as one shot instead of a miniature trailer. I reduced the brief to one subject, one camera move, and one visible change. The reference image defined the starting composition, while the prompt described only the motion.
The simple workflow that helped:
Decide what should visibly change.
Keep one primary camera direction.
Give each reference one job.
Review text, logos, faces, and product details before publishing.
I also found that the reference controls in MiniMax H3 AI Video Generator were more useful when I stopped asking one reference to control everything. A still image held the composition; the motion notes handled pacing and camera direction. Separating those jobs made it easier to see which instruction was causing drift.
The output still needed human review, especially around small text and brand details, but the shorter brief was much easier to iterate on.
When a short AI video feels overloaded, which part do you simplify first: the action, the camera move, or the visual style?


















