Why a storyboard and not a frame grid
A frame grid takes frames at equal steps. For motion graphics, that is not enough. A 300 ms move can fall between two frames, and a long hold can fill half the grid with the same picture.
This tool picks frames from the motion itself:
- Hold: the animation stops on a pose. The tool keeps one frame from the middle of the hold. This frame shows the end state of the move before it.
- Move: the animation is in motion. The tool keeps a frame each time the picture is different enough from the last kept frame.
- Start and End: the first and the last frame of each shot.
What the tool measures
| Value | What it tells the AI |
|---|---|
| Scene start, end, and length | The exact cut points, to the frame |
| Transition in | A hard cut, a gradual change, or a change through black or white |
| Timing | Each move and each hold, in seconds from the scene start |
| Change area | Where in the frame each move happens, and how much of the frame it uses |
| Motion | Static, subtle, moderate, or high |
| Colours | The main hex colours of each scene and of the full video |
The tool decodes small frames, up to 15 each second, and keeps them in memory. A change to a slider updates the storyboard at once, and the tool decodes only the new frames.
What is in the storyboard pack
storyboard-1.pngand more: one row for each scene, with the frames, the times, the motion curve, and the colours.storyboard.md: the prompt. It lists each scene with its timing, then asks the model to recreate the sequence in the tool that you choose.storyboard.json: the same data, for scripts and agents.frames/: each frame as a JPEG file, named by scene, time, and reason.
Tips for a better recreation
- Paste the images before
storyboard.md. - Use one sequence at a time. A clip of 5 to 30 seconds gives the best result.
- Move the frame detail slider up for fast, complex moves. Move it down for slow, simple moves.
- Ask the model to describe each scene first, then to build it. The prompt already asks for this order.
- For an answer about the content of a long video, use Video to AI instead.