How I Write Seedance Prompts: The Borrowed Wounds Rulebook

Put the shot on the page

I use this rulebook when a shot is already conceived and needs to become instructions a video model can follow. During Borrowed Wounds, reducing one working prompt from 2,164 words to 808 meant removing repetition while preserving the physical decisions that mattered.

The task is to describe what the camera can witness: where a body stands, how it moves, what the light reaches and what remains at the end. Use the rules below to revise a draft, then assemble only the blocks your shot needs.

Twelve rules for the working draft

1. Describe the visible alternative

I replace prohibitions about the scene with the physical state I want. In my working tests, negative descriptions sometimes introduced the unwanted object and sometimes removed features too literally, so I do not treat a negative sentence as reliable direction.

Instead of telling people not to look at the camera, I specify their eyeline:

Every man keeps his eyes on the path ahead.

Generation defects are a separate category in this method: terms such as fused digits describe an image failure, rather than an object I am placing in the world. I keep that distinction separate from performance and staging.

2. Count characters, then protect the beginning

I count the complete text before delivery. My recorded web-interface test retained approximately 5,787 characters and silently truncated the remainder; I used 5,600 as a working target, rather than treating that observation as a permanent limit of Seedance itself.

Example from that workflow: the corrected draft measured 4,304 characters. Put decisive instructions in the first two blocks, and shorten texture descriptions, adjectives and repetitions before removing a direct instruction. Check the text actually retained by the interface you are using.

3. Give each action its time

I choose an exact duration and assign the action to intervals. A duration range hands over a decision that belongs to the shot, while a brief action stretched across a longer generation can leave the model filling the remaining time.

This is the timing skeleton used in my rules, with the action still to be filled:

0 to 3 seconds: ...
3 to 6 seconds: ...
6 to 9 seconds: ...

Check that the final interval reaches the requested duration and that the subject still has something physical to do there.

4. Attach units to measurable decisions

I replace relative adjectives with measurements where the shot calls for them. The purpose is to distinguish the subject's movement from the camera's movement, not to fill the page with numbers.

In the recorded revision, people walked at four kilometres per hour while the camera advanced at five, making the overtaking relationship explicit. Camera height changed from 35 to 140 centimetres. These are examples from that shot, not default settings for yours.

5. Start with the scene

The first block is SCENE CONTEXT: what happens, where, when and for how long. I do not begin with a style declaration, a reference inventory or an explanation of what the audience should feel.

Use this opening field order as a check on your own draft:

SCENE CONTEXT
[What happens, where, when, exact duration.]

If the opening cannot identify the physical event, I resolve that before adding optical or surface detail.

6. Give every instruction a home

I keep light, camera, action and performance in their own blocks so that repeated instructions do not compete across the text. A block that contributes nothing is removed altogether.

For example, the direction and temperature of light belong in LIGHTING, while the action's time intervals belong in ACTION. Use the assembly table below as a sorting tool: move each sentence into its proper block before deciding whether it is necessary.

7. Use the field-of-view anchors consistently

My Seedance writing convention uses these field-of-view anchors in degrees: 107, 84, 63, 47, 29, 18, 12 and 8. I use them instead of mixing focal lengths with other optical descriptions.

The field of view holds at 63 degrees from first frame to last.

This is a convention in my working rulebook, not a claim that other angles cannot exist. Check that the movement described elsewhere does not contradict the chosen field of view.

8. Describe the image response

I remove commercial names and named cinematic references from the shot instructions, then describe the qualities I want to see. This keeps the request tied to a visible result.

The recorded replacement for a film-stock name was fine grain, soft highlight transitions, restrained saturation, and earthy greens and browns. Use those qualities only when they belong to your image; the exercise is to replace a name with its intended effect, not to apply the same palette everywhere.

9. Translate emotion into muscle

I write the movement that carries the emotion, rather than asking the model to interpret an abstract label. The rules give lowering the eyes, tightening the jaw and swallowing before an answer as a usable sequence.

Another short fragment fixes a visible state:

Chin level, jaw loose, eyes fixed on the far end of the path.

Read each performance sentence as a camera instruction: if it cannot be photographed, find the physical behaviour that expresses it.

10. Remove commentary addressed to the reader

I cut blocks that explain the purpose of the shot without directing anything visible. The audience's understanding belongs in my decisions about framing and movement.

For example, wanting the audience to recognise the central figure becomes a camera ending close to that person, with the figure occupying the strong position in the frame. The explanation can remain in planning notes; the generation text needs the action that produces it.

11. Make the shot self-contained

I do not rely on a previous shot's description being remembered. Any information needed to produce the current image must be present in the current instructions and references.

The example in the rules is a continuous take written positively:

One continuous shot, the camera holds the take to the last frame.

Check the current scene, subject positions, light and movement directly. A scene number or a summary of earlier events does not supply those physical facts.

12. Let the reference carry appearance

When I attach an image, I keep the accompanying character description short: age, role, current state and one visible trait. The rulebook closes that description with this phrase:

100% matches the reference.

The phrase expresses the instruction, not a guarantee of the result. I remove long appearance passages that compete with the attached image and retain only references that contribute to this shot.

Assemble the shot

Final pass before delivery

• [ ] The first frame contains the subject and a readable action.
• [ ] Left and right are described from the camera's position.
• [ ] The subject's gaze and direction of travel are clear.
• [ ] Light has a source and direction, with atmosphere specified when needed.
• [ ] The action reaches the final frame rather than stopping early.
• [ ] Reference descriptions support the images instead of competing with them.
• [ ] The complete text has been counted, and decisive instructions remain near the beginning.

Keep the physical decisions

I shorten a prompt until its remaining words carry the shot's movement, performance and continuity. The finished draft should make those decisions easier to inspect before generation, and easier to compare with the result afterwards.