Looksmaxxing for motion graphics. An agent skill that turns AI slop videos into films that look made by a motion designer.
Where this skill's taste comes from, and what it does not contain
The question
The taste in this skill was learned by studying launch films other people made and published. Is it right to turn that into a skill?
The answer this skill takes
Learn the way designers learn: by studying published work and writing down principles. Never redistribute the work, and never clone it.
- The films stay private study. About 50 designer-made launch films (from public social posts and brand channels) and about 30 AI-made films were watched closely by the skill's authors: what was decided, why it works, what fails. They live only in the authors' private study folder. No frame, clip, audio, logo or screenshot from any of them is part of this skill.
- What ships is principles, in our own words.
references/judgment.mddescribes decisions ("the logo became the product's own control"), not compositions. Examples are described generically (for instance "a launchpad's pill-shaped icon"), so the skill passes on reasoning rather than anyone's look or brand. Ideas and principles are how craft has always been taught: critique, essays and design school, not copying. - The pictures are original. The calibration strips (
judgment-studies-*.jpg) are new images generated for this skill with the Codex CLI's built-in image tool, from the prompts inplates.json/plates2.json(kept inREF/motion-eye/studies/src/). Every brand in them is invented. No pair re-creates a specific reference frame. Each pair shows a judgment, and the pairs are deliberately drawn in different styles so they don't become a look to copy. - The skill forbids cloning. It tells the model to derive the look from the subject's own brand, never from a reference, a previous film or the strips.
- The user's own material is the user's.
taste/verdicts.mdrecords what this user said about films that were made for them.
Why not a "gold strip" of reference frames?
Two reasons, one ethical and one practical:
- Redistribution. Frames cut from other studios' films and shipped inside a tool are their work, not ours.
- Homogenisation. An earlier version of this skill used a gold gallery (first real frames, then image-generated re-designs of specific reference frames). Agents copied it. Films for eight different brands came out looking like one designer's reel: a giant cropped word, letters filled with texture, a colour-flood wipe, a dot that becomes the logo. A strip of looks teaches a look. A strip of judgments, drawn in deliberately different styles, teaches judgment.
Regenerating
The prompts and the strip builder (plates.json, plates2.json, make_strip.py, raw plates) were kept in the author's working folder and are not part of this repo. Run the prompts through any image model (scripts/imagegen.py refuses prompts that ask for text, logos or UI, so any strip prompt with lettering needs a different image tool), name results <NN>a-default.png / <NN>b-decision.png, run make_strip.py. Nothing in this skill depends on regenerating them.