I’m leony, an indie maker, and I’ve been collecting Seedance experiments because a vague video idea is not always a useful video prompt. “A cinematic shot of a city at night” describes a mood, but it leaves many production decisions unstated. What is the subject doing? Where does the camera move? What changes halfway through the shot? What should the model avoid adding?
That led me to build the Soutine Seedance Prompt Library, a small collection of 66 prompts with real clip previews. It is meant as a practical reference for studying shot language, not as a list of guaranteed formulas.
From idea to describable shot
A usable video prompt usually gives the model a few connected pieces of information:
- Subject and action: who or what is visible, and what happens during the shot.
- Camera movement: whether the camera pans, tracks, pushes in, cranes, or stays locked off.
- Timing or a mid-shot beat: what changes after the opening moment, such as a subject turning, a light switching on, or an object entering frame.
- Aspect ratio and framing: the intended shape and a useful description of the shot size.
- Constraints: details such as “no text,” “no logos,” or a clean, stable background when those matter.
These details do not remove the creative part. They make the creative intention easier to test. A prompt that names a camera move and a beat gives me something concrete to compare when I revise the clip.
Preview, copy, adapt
The workflow behind the collection is intentionally low-friction. I preview a clip, read the exact wording, copy it, and adapt one variable for the next iteration. If the camera motion works but the subject feels static, I change the action rather than replacing the entire prompt. If the action is good but the result becomes visually confusing, I add a clearer framing or a constraint.
Small changes are useful because video generation has many moving parts. Changing camera movement, timing, subject behavior, and visual style simultaneously makes it hard to know what caused an improvement—or a failure. Keeping most of the prompt stable turns each attempt into a small experiment.
The working collection is available at the Soutine Seedance Prompt Library, and the small companion reference package is soutine-seedance-prompts on GitHub. Browsing and copying the examples is free. Creating new generations in Soutine consumes Soutine credits, so the library is best thought of as a reference and starting point rather than a promise of unlimited generation.
A prompt I would improve first
Suppose a baseline says: “A paper boat floats through a rainy street.” I might revise it to: “A close tracking shot follows a paper boat drifting along a rain-filled street gutter; halfway through, it passes beneath a warm shop light; 16:9, gentle handheld motion, no text or logos.” The second version gives the shot a subject, action, camera behavior, timing beat, format, and constraint without dictating every pixel.
If you have better ways to describe camera language, timing, or stable motion, I’d be glad to see those ideas applied to the collection. The most useful prompts are the ones that make the next experiment easier to understand.
Top comments (0)