What Do First and Last Frames Mean?
The first frame freezes the instant a shot begins; the last frame freezes the instant it ends. They are not two independent visual ideas. They project the start and end boundaries of the same shot. Character position, held objects, orientation, and visibility must agree with the corresponding boundary.
Providing both frames delegates part of the motion between them to the generation system. They supply endpoint constraints rather than proof that either generated endpoint is correct. When exact intermediate action matters, specify and inspect that action too.
When Should You Use First and Last Frames?
Ask two questions first:
- Can one static image express the end state clearly? For example, an umbrella changes from closed to open, or a character changes from standing to seated.
- May the path between the two states vary? If the precise turn does not affect the story, the method may fit. If a hand must seize evidence at the instant a particular line is spoken, the path itself is dramatic action.
| Shot | Details |
|---|---|
| Paper lanterns in an empty courtyard change from dark to lit | Judgment: Suitable; Reason: Both endpoints are clear; the lighting sequence is not central |
| A girl passes a letter from her sleeve to someone inside a doorway | Judgment: Use caution; Reason: Contact, receipt, and concealment order affect continuity |
| Teacups collide and tip, and spilled tea reveals hidden writing on paper | Judgment: Do not rely on first and last frames alone; Reason: The collision and appearance of the writing are the information reveal |
| A character moves from a window and sits at a table | Judgment: Suitable; Reason: The endpoint composition matters; the walking path may vary |
How to Write the First Frame, Last Frame, and Motion Description
Shot purpose: What must the audience understand at the end?
First frame: Include only visible start-state facts—character position, posture, held objects, gaze, foreground, middle ground, background, light source, and composition.
Last frame: Include only visible end-state facts, matching the shot’s endpoint.
Must happen in between: List only indispensable action order, dialogue timing, or camera behavior.
Preserve: Keep identity, clothing, scene direction, key props, and light sources unchanged.
If “Must happen in between” contains several contacts and reactions that cannot be lost, the shot should not be entrusted wholly to interpolation. Divide it into clearer boundaries.
Original Example: Opening an Umbrella on a Rainy Night
Purpose: after the umbrella giver leaves, Aunt Zhou chooses to follow and steps outside. She has not caught up. The accepted script and storyboard establish these boundaries. This original design has not been generated.
Start frame: medium shot under a narrow alley’s eaves. Zhou stands one step inside, holding a closed dark-blue umbrella in her right hand, its tip on the ground. Rain lies screen right; warm door light comes from behind left. She looks into the empty alley. The umbrella is closed.
End frame: same camera and light. Zhou stands outside the threshold, right hand still on the handle, umbrella fully open overhead. Body and gaze face the alley entrance; the door stays open. This projects the accepted end without adding the giver or a completed reunion.
Motion: Zhou lifts the closed umbrella, opens it in her inside position where sufficient room is already established, then crosses the threshold and settles facing the alley. Fixed camera; identity, outfit, door and light remain. End with right hand on handle, open umbrella overhead and Zhou outside, ready for the next shot.
Viewers need the choice to leave, not an exact umbrella-opening technique. If a hidden key falls from the ribs and someone steps on it, the path reveals information and needs explicit shot coverage rather than just two pictures.
A correct input end frame does not prove generated opening and threshold crossing. Inspect actual intermediate action and final frame; this design is not a passed video.
Common First-and-Last-Frame Failures and Fixes
| Issue | Details |
|---|---|
| Start frame shows spilled water | Effect: End appears early; Fix: Keep only start facts |
| Held object differs from accepted end | Effect: Two end states compete; Fix: Use the shot’s accepted end |
| Large gap between frames | Effect: Motion path too free; Fix: Reduce change or split shot |
| Key action only lies between frames | Effect: Action may be missed; Fix: Give it clear shot coverage |
| Each frame uses its own layout | Effect: Axis and light jump; Fix: Use shared space, direction and light |
| Still says “first, then, afterward” | Effect: One image carries a sequence; Fix: Freeze one instant |
The input end frame does not certify the output
| Check | Rain-scene evidence |
|---|---|
| Start | Closed umbrella; Zhou inside, no premature end |
| Middle | Same Zhou opens umbrella and crosses, without teleportation |
| End | Outside, open umbrella, right hand on handle, intended direction |
| Continuity | Identity, outfit, door and lighting match next shot |
If output ends inside or switches the handle to another hand, repair against the accepted end rather than substitute the input picture for observation. If a different ending is acceptable, confirm it in script and storyboard and update the next shot, instead of creating a separate ending only in prompt text.
FAQ
What if the motion between the first and last frames becomes unstable?
First decide whether the intermediate path matters. If it does not, reduce the difference between the two frames and strengthen the identity, spatial, and lighting constraints. If it does, split the key action into separate shots or specify the action order instead of asking two images to imply it.
Does every shot need both a first and a last frame?
No. One keyframe that defines the start state is often enough. Add a last frame only when the production process genuinely needs to control the endpoint and the shot is suitable for interpolation.
Do it with the skill
In Claude Code, say: “Use /short-drama-storyboard to decide whether this shot suits first-and-last-frame generation. Write separate static descriptions containing only the start and end states, then identify any intermediate action that must be explicit or should become another shot.” In Codex, write $short-drama-storyboard.
Read the method: Start/end boundaries, interpolation and end checks · Motion state and delivery
About Drama SkillsThe short-drama-storyboard skill on GitHub