Requires
A running Scrollmark editor whose commands you can call, signed in to SocialGPT: motion graphics are generated on SocialGPT's servers and cost credits. Without it, the rules for when a graphic is earned still hold in any editor.
A motion graphic is a short designed clip for one moment: a photo cutaway, a stat, a name tag, a step card, a product or price tag, a notification, a headline card. A motion designer builds it from the footage, the words spoken there and the project's design style, and the editor places it on the timeline when it is done.
1. Only when asked
Make a graphic only when the user asked for one, in any words ("add a title", "put her name up", "show the price", "make the stat pop"). Never add a hook, title or text graphic on your own, even to open a video or after a recut. Captions are not motion graphics: the editor captions the whole video (editor-captions).
The template commands graphics.addTitleCard, graphics.addCTA, graphics.addText and packs.insertTitleScene are turned off; titles, hooks, CTAs and text graphics all come from media_generate with kind: "motion".
2. One call per moment
Call media_generate with kind: "motion", one call for each moment. Several moments ("a name tag and the price") are several calls in the same round.
at_s: where the moment starts on the timeline, at the line it belongs to. Readfootage_getTranscriptortimeline_describeand start a few tenths of a second before the key word. A title or hook for the opening goes at 0.duration_s: as long as the line it sits on, usually 3–6 s; at most 30.layout:overlay(default) sits over the footage and leaves the speaker in view; use it for name tags, stats, product tags, steps and notifications.full-framecuts away to a designed card while the voice continues; use it for a headline or title card, a correction, or a list that needs the whole frame.prompt: what the moment is and the words it shows, in a sentence or two ("Name tag: Dr Maya Chen, sleep researcher"; "Price drop: was $32, now $24, this week only"). Leave out fonts, colours and animation: the design style and the designer decide those.exact_strings: every name, handle, URL, number and price exactly as it must appear, from the user or the project, never re-spelled from the transcript.images: only when the user wants a photo or logo in it, as slot names to project image ids ({"product": "<mediaId>"}). Without images the designer works with type and shapes; do not generate an image first unless they asked for one.
3. While it runs
It is non-blocking and takes a few minutes. Say what you started ("a name tag for Maya at 0:04"), finish the rest of the request, and end the turn; do not say it is on the timeline. When a [generation finished] message gives its elementId, it is already placed: do not place it again or call graphics.addCode. Confirm once in one line.
4. Changing one
- Move it with
timeline.moveClip, lengthen or shorten it withtimeline.setClipDuration, or remove it withtimeline_deleteClips, using itselementId. - Different words, a different look or a fix inside the graphic: generate it again with the change in
prompt, then delete the old clip once the new one is placed. Say that it makes a new graphic.
5. Say what you did
Name each graphic, where it sits and what it shows. If the user asked for something that isn't a moment graphic (running captions, a whole animated video), say what you did instead.