visuals-core
Skill0 starsChanged 21 days agoMITView on GitHub
Shared tools and the shooting guide for the /visuals effect skills (orbit, zooming, words-from-behind, icon-on-palm, zig-zag-graph, paper, thanos-snap, share-comment). Use it to plan or film a laptop-on-a-desk reel for these effects, to measure coordinates (grid.py), to build the laptop mask (mask.py), or to stitch the rendered effect clips together.
- Installs packages
In this repository8 other files it gives its agents.
- icon-on-palmSkill
- orbitSkill
- paperSkill
- share-commentSkill
- thanos-snapSkill
- words-from-behindSkill
- zig-zag-graphSkill
- zoomingSkill
The file
We split this skill into the rules it gives an agent
4 rules. About 0 are common, 1 is distinctive, and 3 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈1
Distinctive · estimate
This author's own choices. Worth a look.
3
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 18
Keep the folders side by side: Every effect script imports ../visuals-core/vfx.py, so install all 9 folders into the same skills directory.
- Distinctive: line 23
Find the moment: Ask the user for the clip and the rough time of each beat, or scrub it with ffmpeg stills.
- Very specific: line 25
Measure: Run grid.py on a still from the beat and read coordinates off the grid.
- Very specific: line 32
Preview: Every effect takes --preview 0.3,1.0,2.0 (effect-local seconds) and writes PNG stills in seconds instead of a video.
≈ “Common” is an estimate: at least 1 in 5 of the 364,980 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
visuals-core
The engine under the 8 /visuals skills. Each effect skill is a Python script that turns a few seconds of your own
footage into a finished effect clip (MP4 with the original audio). This folder holds what they share:
vfx.py: ffmpeg video I/O, laptop/screen/hand masks, easing and springs, RGBA compositing, sticker textgrid.py: writes a still with a labelled pixel grid, so you can read off coordinates for flags like--targetmask.py: buildslaptop_mask.png, the cut-out every "behind the lid" effect uses, plus a red check overlay
Requirements: Python 3.9+, pip install opencv-python numpy pillow, and ffmpeg on PATH. SVG logos are drawn
with headless Chrome/Edge if installed (or pip install cairosvg); PNG always works.
Keep the folders side by side. Every effect script imports ../visuals-core/vfx.py, so install all 9 folders
into the same skills directory.
The workflow (same for every effect)
- Find the moment. Ask the user for the clip and the rough time of each beat, or scrub it with ffmpeg stills. Times are in seconds of the source clip.
- Measure. Run
grid.pyon a still from the beat and read coordinates off the grid. Look at the image; do not guess.
Openpython <skills>/visuals-core/grid.py --in clip.mp4 --t 30.9 --out grid.jpg python <skills>/visuals-core/mask.py --in clip.mp4 --t 30.9 --out laptop_mask.png # hand-free framelaptop_mask_check.jpg: the red area must cover the whole laptop and nothing else. If it spills onto a dark table, raise--min-yor lower--v-max. - Preview. Every effect takes
--preview 0.3,1.0,2.0(effect-local seconds) and writes PNG stills in seconds instead of a video. Look at them and adjust flags. - Render. Drop
--preview, set--out effect.mp4. Output is 1080x1920 30 fps H.264 + the clip's audio. - Stitch. Concatenate in your editor, or:
printf "file 'orbit.mp4'\nfile 'zoom.mp4'\n" > list.txt ffmpeg -f concat -safe 0 -i list.txt -c copy reel.mp4
Flags every effect has
| flag | meaning |
--in, --start, --dur, --out | source clip, effect start (s), effect length (s), output |
--size 1080x1920, --fps 30 | output format (portrait by default) |
--mask laptop_mask.png | the laptop cut-out (auto-detected if omitted) |
--preview t1,t2,... | write stills at these effect-local times instead of rendering |
--gamma 0.86 | brighten dark footage (lifts the mids; 1.0 = off) |
--wb 1.045,0.995,1 | white-balance gains B,G,R (this example cools a yellow wall) |
--no-audio | do not copy the source audio |
Shooting guide (what makes these effects possible)
- Locked-off camera. Phone on a tripod, vertical, never touched. The hand and lid tricks key against a still frame.
- Framing. Laptop in the lower ~55% of the frame, plain wall above it: orbit, words, graph and the sticker live there.
- Even front light, exposure and focus locked (long-press AE/AF lock). 4K if you can:
/zoomingdives deeper. - Big screen text. Zoom the app in (Ctrl/Cmd
+), dark or light mode, notifications off. - Hand-free moments. Hold still with hands off the laptop for 2-3 s now and then. Those frames are the
--ref-timethe hand effects key against. - Clean plate for
/thanos-snap: after the take, lift the laptop out and record 5 s of the empty desk without moving the phone. - Per beat: say the line, type the slash command slowly, Enter, then do the action:
| effect | what to do on camera |
/orbit | hands off the keyboard, hold still 3-4 s |
/zooming | hands still 4 s (choose a target on screen: an icon, a mascot's eye) |
/words-from-behind | hands still 3 s |
/icon-on-palm | raise an open palm, flat and facing up, beside the laptop; hold 4 s, move it only slightly |
/zig-zag-graph | one index finger draws a rising zig-zag left to right above the laptop (~2 s) |
/paper | hands on the keyboard, mostly still, 3-4 s |
/thanos-snap | hand up beside the laptop, one clear snap, hold 1 s, then hands out of frame |
/share-comment | 4 s of the empty desk (or use the clean plate as the background) |
Full guide with the reasons behind each rule: docs/SHOOTING.md in the repository.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

