aievolutionpl/social-motion-agent

Motion Effect Picker

Source: tools/repos/awesome-ai-motion (MIT; GSAP bundled under its no-charge licence, see its ATTRIBUTIONS.md). 637 effect cards, 128 of them with a rendered clip and HTML template, 11 recipes, decision tables. Never read the whole index; query it.

Steps

1. State the scene intent and medium in one sentence ("the question that opens the video, shorts").
2. Query:

python -S studio.py motion find reveal --purpose attention --medium shorts --rendered --limit 8
python -S studio.py motion show mask-reveal
python -S studio.py motion recipe            # lista receptur
python -S studio.py motion recipe edu-hook   # kroki z czasem i parametrami

Purposes: attention, explanation, comparison, sequence, emphasis, transition, data, feedback, atmosphere, branding. Media: explainer, shorts, scrolldeck, presentation, product-demo, data-story, web-ui. Also --family, --level. 3. Choose 2 or 3 candidates, read each card's use_when, avoid, params, ease, pairs_with. Take the card defaults first, then adjust. 4. Compose with the shared rules: one focus per scene, one main action plus one or two supporting, the finished state held long enough to read, transform and opacity only. 5. Implement as a pure function of time. The card snippet is GSAP; translate it, do not import it:

• seconds to progress: p = clamp((t - start) / duration); stagger = start + i * gap
• easing from templates/motion-helpers.js (easeOut, easeIn, easeInOut, backOut, elasticOut, spring): power3.out is easeOut(3)
• yPercent: 110 -> 0 is translateY(110% -> 0%) on the line inside an overflow: hidden window

6. Build with motion-ui-polish (tokens and gate), view the frames at start, main action and end state, compare with the card's good_example.

When a template exists

If show prints html_template, open it to see structure and timing, copy it into the job's source/, and replace text or data. Keep the timeline. Do not run the catalogue's render.mjs without installing its Playwright dependency; our pipeline renders through vstudio or the HTML renderer.

First five seconds (educational and explainer videos)

0 to 1 s a question or surprising result; 1 to 3 s a clue to the mechanism or comparison; 3 to 5 s the learning promise with a readable hold. One claim per screen. Large camera moves only to reveal the question or the result. Start from motion recipe edu-hook, then mask-reveal, split-compare, count-up, annotation-callout.

Limits

• Card parameter names are partly Korean; values, ranges, ease and snippet are language neutral. Descriptions and prompts come back in English.
• A planned card has no clip or template; use its parameters and prompt only.
• Overused effects look generated. Do not repeat one effect on more than one element per scene, and do not stack three effects on the same moment.