Global-AI-Group-Inc/content-generation-skills / skills/mira-after-effects
mira-after-effects
VerifiedBuild motion pieces inside the user's own After Effects through the Mira MCP: read the project, mark the beat of the music, animate with the Mira Motion Kit (kinetic type, camera depth, velocity-matched transitions, expressions, grade) or ExtendScript, add plates, overlays, templates and sound, check contact sheets, render on the user's machine. Use it when the user says "build this in After Effects", "собери ролик в афтере", "промо в After Effects", "make a motion reel", "add grain and a LUT", or when adobe_status reports a connected After Effects. Requires the Mira for Adobe panel with the bridge switched on. NOT for generating clips (mira-generate) or Blender scenes (mira-blender-scene).
| File | Kind | Size |
|---|---|---|
| SKILL.md | skill md | 9.5 KB |
| references/bank-schema.md | markdown | 2.9 KB |
| references/depth.md | markdown | 4.5 KB |
| references/doctrine.md | markdown | 5.2 KB |
| references/motion-kit.md | markdown | 16.5 KB |
| references/recipes.md | markdown | 7.1 KB |
| references/workflow.md | markdown | 5.3 KB |
references/doctrine.md
View on GitHub# Why a scripted motion piece looks flat, and what makes it rich
A composition built only from solids, text layers and position keyframes reads as a slide deck
with movement. Nothing in it has ever been photographed, so nothing carries light, grain or
depth. The fixes below are the ones a motion designer applies by habit; apply them in this order
and stop when the piece holds together, not when every trick has been used.
## 1. Depth: three to five planes, never one
Every scene has a plate, a texture on it, the subject, something in front of the subject and a
grade over the lot. A plate is a gradient clip, a smoke plate, a blurred still or a full-bleed
video, never a flat colour on its own. The front plane is a light leak, dust, a lens smear or a
foreground shape with heavy blur. Scale contrast does the rest: one giant word against one small
caption, one huge mockup against a row of tiny chips. Parallax comes free from a 3D camera with
a slow push and layers at different Z, or from position keys that move background and subject at
different speeds. How far apart the planes move, how much haze each gets, push-ins, grounding
shadows and the speed budget for motion blur are in `references/depth.md`.
## 2. Texture: grain, halftone, dirt
Clean pixels are the strongest tell of a render. Put film grain over everything at 25-40%
(Overlay or Screen), a halftone or scanline plate on flat colour at 15-30%, dirt or dust on
dark scenes. One texture per plane; two textures on the same plane look like a mistake.
## 3. Grade and vignette
One LUT for the whole piece, chosen before the first scene and never changed mid-way; footage
is pre-graded before import so type and UI stay clean. A vignette of 20-35% closes the frame.
Chromatic aberration of 2-4 px on the very edge of a fast move sells speed; on a still frame it
is noise.
## 4. Light
Something in every scene should glow or leak: a light leak on the cut, a glow behind the hero
word, a soft orb behind a logo, a moving highlight across a chrome title. Light is what the
audience reads as expensive.
## 5. A typography system
Two typefaces, three sizes, one alignment rule. Display type is huge and set tight; body type is
small and airy. Kinetic type moves on the beat and in one direction per phrase: slide, rise or
slam, never all three. Type entering a frame should enter over something, not onto emptiness.
## 6. Cuts on the beat, with sound
Every cut sits on a beat. A whoosh starts 0.3-0.5 s before the cut and its transient lands on
it; a hit lands on the peak; a riser runs the 1-2 s before a drop. Transitions are footage
(light leak, film burn, luma wipe), not opacity fades. A fade is reserved for the very end.
## 7. Framing devices
A phone, a screen, an old TV or a billboard mockup gives footage a place in the world. The
footage plays inside the device; the device moves, not the footage.
## 8. Restraint
One hero per shot. At most three moving things per second. One transition family per piece.
If a scene needs a second texture or a third font to work, the scene is wrong, not the toolbox.
## 9. Against the default look
What every generated motion piece does by default, and what to do instead:
- Centred text on a full-frame gradient with fades → a plate with depth, type anchored to an edge
or a grid, entrances that are verbs (slide, mask reveal, slam) and a real transition on the cut.
- Full-frame linear gradients on dark → they band in H.264; use a textured plate or a radial glow.
- Inter, Roboto, Poppins, the same weight everywhere → a display face with character and a quiet
text face; contrast in weight (300 against 800-900) and tight tracking on the big lines.
- Every move with the same ease → at most two eases per scene; entrances longer than exits
(≈0.4 s in, 0.25 s out); the slowest scene about three times the pace of the fastest.
- A scene that builds and then just sits → build, breathe, resolve (≈30 / 40 / 30 % of the scene).
- Motion blur on everything → only on the snaps (whips, slams, zooms), never on text being read.
- A picture dropped in flat → tilt it in perspective, give it a slow push (100 → 104 %) or put it
in a device frame.
## 10. Platforms
| Platform | Frame | Keep clear | Length |
|---|---|---|---|
| Reels | 1080×1920 | bottom ~20 % (caption, buttons), right ~12 % (actions), top ~10 % | hook in the first 1-2 s |
| TikTok | 1080×1920 | bottom ~20 %, right ~15 %, top ~8 % | hook in the first 1-2 s, loop the end into the start |
| Shorts | 1080×1920 | bottom ~15 %, right ~12 % | hook in the first 2 s |
| YouTube | 1920×1080 | lower third while the player bar shows, top-right corner | title card in the first 5 s |
Loudness about -14 LUFS integrated with the peaks under -1 dBTP; the voice sits above the music.
`adobe_render` takes the platform as `preset` and reports a composition of the wrong shape.
## Check before rendering
- Five planes in the busiest scene, three in the quietest.
- Grain and vignette on the whole piece, one LUT.
- Every cut on a beat with a sound under it.
- Two typefaces, no more; type never enters onto an empty frame.
- Something glows in every scene.
- Nothing from the default look (section 9); nothing important in the platform's UI zones (section 10).
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| reverse-engineer-anythingmorluto/reaReverse engineer native, managed, Electron/JavaScript, packaged, firmware, and browser targets with REA, and build agent environments (AgentEnv bundles) of web applications from a live instance or a capture, and of Linux desktop apps with a task from a request. Use shipped-artifact or requested runtime evidence to explain features, compare versions, decompile code, or guide a reconstruction. Skip REA for ordinary source-repository architecture analysis. | Stars: 472k | Updated | Claude CodeCodex | Verified | Install |
| diagnosing-superpowersobra/superpowersUse when a superpowers session went wrong and your human partner wants to know why — repeated work, ignored plans, stumbles, poor results, a skill that didn't fire, "it took too long", "why is it so expensive", "what is it doing" — or wants to build a bug report for the superpowers maintainers, for the current session or a past one identified by id or path, on any harness. | Stars: 323k | Updated | Claude CodeCodex | Verified | Install |
| dispatching-parallel-agentsobra/superpowersUse when facing 2+ independent tasks that can be worked on without shared state or sequential dependencies | Stars: 323k | Updated | Claude CodeCodex | Verified | Install |
| executing-plansobra/superpowersUse when executing an implementation plan in the current session as the implementer yourself — your human partner chose inline execution, or no subagent tool is available | Stars: 323k | Updated | Claude CodeCodex | Verified | Install |