A lot of inconsistent motion comes from prompts that ask for a mood instead of a sequence. When you only describe the vibe, the model has to invent the mechanics. That is where drift starts.
A better structure is to separate the prompt into four parts: subject, camera, environment, and timing. The subject defines who or what the clip follows. The camera defines how the frame behaves. The environment defines what stays anchored. Timing describes how movement unfolds across the shot.
A practical pattern
Start with a short subject line, add a camera instruction that uses one movement verb, then lock the setting with one or two environmental anchors. End with a timing phrase such as slow reveal, accelerating push-in, or steady lateral drift. This creates a prompt that is easier to tune one variable at a time.
If the result still feels unstable, remove adjectives before adding more. Excess styling language usually muddies the directional cues you actually care about.
No comments yet