Cepter EFX is an After Effects panel that bundles the things AMV/video editors reach for constantly into one place: offline AI captions, offline voice/stem isolation, beat detection, a project-organising tool, precomposing, animation copy/paste, custom bezier easing, anime "dead frame" cleanup, and batch render-queueing.

Everything that can run offline does. Nothing is uploaded anywhere, there is no account, and the panel never phones home for anything except the optional one-time downloads of its AI engines/models (which you trigger yourself, from inside the panel).

Every action in every tab is a single undo step: one Ctrl/Cmd+Z reverses the whole operation, not just its last keyframe or layer.

• Adobe After Effects 2019 (16.0) or newer. The panel declares compatibility up to any future version ([16.0, 99.9] in its manifest).
• Windows. The installer is for Windows, and the AI and FFmpeg tools (Auto Captions, Voice Isolation, Dead Frames and the accurate Beat Markers analysis) use Windows engines. The panel can be installed on macOS by hand, but those tools won't run there.
• No Adobe Creative Cloud app required. The panel and its installer both work with a standalone After Effects install.
• Nothing else to pre-install. Auto Captions, Voice Isolation, Dead Frames and Beat Markers all download the small helper tools they need (an audio decoder, and per-feature AI engines/models) from inside the panel, on demand, the first time you use them.

You'll be given a Cepter_EFX_<version>_Installer.zip. Unzip it anywhere: inside you'll find the signed .zxp and a short instructions file.

One-time setup: ZXP/UXP Installer

If you've installed Cepter EFX (or any other .zxp/.ccx extension) before, skip this: you already have it.

1. Download aescripts' free ZXP/UXP Installer for Windows: aescripts.com/learn/zxp-installer/
2. Run its installer and finish setup. It's a small standalone app: it does not need Creative Cloud installed.

Installing

1. Close After Effects if it's open.
2. Open ZXP/UXP Installer.
3. Drag Cepter_EFX_<version>.zxp onto its window, or use File → Open (Ctrl+O) and pick the file.
4. It shows its own success message once done.

Opening the panel

1. Start After Effects (or, if it was already running, fully quit and reopen it: extensions are only scanned for on startup).
2. Window → Extensions → Cepter EFX by Intrcepth.
3. The panel opens as a normal dockable panel: dock it wherever you like.

Panel size

The panel opens at 340×700 by default, and can be resized (docked or floating) anywhere between 290×420 and 600×1400.

Troubleshooting install

• ZXP/UXP Installer won't detect After Effects, or fails. It has its own "fallback custom install" mode for troubleshooting: check aescripts' FAQ before using it: aescripts.com/knowledgebase/index/view/faq/zxp-installer-faq/
• It installed but doesn't appear in the Extensions menu. Fully quit After Effects (not just close the project) and reopen: CEP extensions are only scanned on startup.
• Still stuck? intercepth.dev has every link, including where to ask for help.

There is no in-panel updater, and the panel never checks a server. Older builds eventually stop opening until you install a current one, so it's worth updating whenever a new version is out.

Updating itself: install the new .zxp over the old one the same way you installed it the first time. Nothing needs uninstalling first, and your settings, theme colour, and presets all carry over untouched.

If you're current, you'll never see an update prompt.

Welcome screen

The very first time you open the panel, you land on a welcome screen showing the Cepter EFX logo, "Made by Intrcepth", the current build/date, a link to the tutorials Discord, and a Let's Work Together button. Click it once and you go straight to the main app on every future launch: the welcome screen isn't shown again.

Layout

From top to bottom, the panel always shows:

1. Quick Create: an always-visible toolbar (anchor/align pad + three quick-create buttons) that's available no matter which tab you're on.
2. The tab bar: nine icon buttons for the panel's tools (Tools, KeyframeFX, Graph, Captions, Voice Isolation, Beat Markers, Precomp, Dead Frames, About & Updates).
3. The active tab's content.

Tooltips

Hover almost anything for about half a second and a tooltip explains exactly what it does.

Panel-wide behavior

• Every long-running job runs in the background. After Effects is never polled on a timer while Auto Captions, Voice Isolation, Beat Markers or Dead Frames are working: clicking things elsewhere in the panel stays responsive.
• Hold Shift on several Quick Create buttons to get a per-layer result instead of one combined result (see below): a small activity dot appears in the app title while After Effects is working through queued requests; it clears on its own.
• F5 reloads the whole panel. Escape closes any open dropdown.

Settings persist automatically

Every toggle, slider, colour, and last-used value you set anywhere in the panel is saved the instant you change it and restored on your next launch: there's no "Save Settings" button anywhere, and nothing needs to be redone after restarting After Effects.

Sits above the tab bar and is available regardless of which tab is open.

Anchor / Align pad

A 9-point grid (corners, edges, centre) plus a two-way toggle above it:

• ANCHOR (default): sets the anchor point of every selected layer to that position. The layer's position is compensated automatically so nothing visually shifts.
• ALIGN: aligns selected layers to that position on the composition instead.

Hold Shift while clicking a pad button to constrain to one axis: the top/bottom-middle buttons move only vertically; the left/right-middle buttons move only horizontally. The four diagonal buttons have no single-axis version of this (a corner is inherently both axes), so while Shift is held they visibly grey out and stop responding to clicks: this applies in both Anchor and Align mode.

NULL OBJECT

• Nothing selected: creates one null spanning the whole composition.
• Layers selected: creates one null spanning the full selection, parented above all of them, and parents every selected layer to it.
• Hold Shift: instead of one shared null, every selected layer gets its own null, spanning just that layer, named after it, parented only to it.

ADJUSTMENT

Same placement/duration logic as Null Object, but no parenting happens: it's just an adjustment layer over the selection (or the whole comp). Hold Shift for one adjustment layer per selected layer.

SOLID LAYER

Same again, filled with whatever colour is currently in the swatch next to the button. Hold Shift for one solid per selected layer.

• Swatch button: opens the shared colour picker (see §16). Your last-chosen colour is always remembered between sessions. Default: white.

(tab icon tooltip: "Tools: quick effects & precompose")

A grid of quick one-click tools, plus a single-button project organiser.

PRECOMP SEP

Precomposes every selected layer into its own comp, numbered by timeline order, filed into a project folder called "Master Comps", and named "<LayerName> Precomp". Each new precomp is trimmed to that layer's own in/out span and reseated on the timeline so nothing visually shifts.

• Shift+click to run Un-Precomp instead: select a precomp layer and everything inside it comes back out into the current comp, retimed and trimmed so nothing jumps, including that precomp layer's own stretch/trim, and the parenting between the layers that were inside it. If the precomp layer had been moved, scaled or rotated, a Controller null takes its place carrying that transform, and the extracted layers are parented to it instead of being left unadjusted. What it can't bring along is named explicitly in the result rather than silently dropped: effects, masks, blend mode and opacity that lived on the precomp layer itself (parenting doesn't inherit these), and a precomp layer using Time Remap is skipped outright: that timing can't be flattened without changing it.
• There is no custom-name option: precomps are always auto-named after the source layer.
• If Safety Engine is on (see §15), the whole comp is duplicated to a comp named "Safe Master" before anything changes.
• One undo step for the whole operation.

FILL

Adds a Fill effect to every selected layer, using whatever colour is currently in the Quick Create swatch.

FIT TO COMP / FIT WIDTH / FIT HEIGHT

Scales and centres the selection to exactly fill the composition (Fit to Comp), or just match its width (Fit Width) or height (Fit Height). Needs After Effects 24.x or newer: the underlying AE scripting API this relies on doesn't exist in older versions.

PURGE

Clears all of After Effects' caches: RAM preview and image caches, in one click. Use it when AE gets sluggish.

ORGANISE PROJECT

A single accent button below the tool grid: files every loose item at the top level of your Project panel into named folders, based on what each item actually is (read off the item itself, not guessed from its name):

Comps · Precomps · Video · Images · Sequences · Audio · Design · Solids · Placeholders · Other

• Only the folders that actually get something are created.
• Names are never changed: renaming is what breaks expressions and file links, and nothing here needs it.
• Only items sitting loose (not already filed into a folder by hand) are touched: anything you've already organised is left exactly where it is.
• A comp nested inside another comp is classed as a Precomp; a top-level comp goes to Comps. An image file with a duration (an image sequence) is filed as footage in Sequences, not among stills.
• One click does the whole thing: under the hood it silently scans first, and if there's genuinely nothing loose to sort it tells you so instead of doing anything ("Everything loose is already filed: nothing to do."). One undo step either way.

(tab icon tooltip: "KeyframeFX: copy & paste animation")

Copies an animation source: either a layer you select, or an .ffx preset, and pastes it onto as many target layers as you like. Only one source is active at a time: picking a preset clears a copied layer, and copying a layer clears a chosen preset, so a press only ever does the one thing.

What gets copied from a layer, and only this

Selecting a layer as the source copies exactly three things: its effects (full parameter values, not just the animated ones), its animated Transform, and its animated Audio Levels. That's the whole contract: nothing else is ever read, so nothing else can ever be written onto a target.

Everything else on your target layers is left completely alone: text, masks, shape contents, layer styles, material options, Time Remap, markers, and every layer switch (blend mode, motion blur, 3D, quality, sampling, collapse, auto-orient, guide, adjustment). Copy a text layer, paste onto five others, and all five keep their own wording: they only gain the animation. A paste adds animation; it never rewrites a target's content.

If the source copy carried something AE's scripting API genuinely can't recreate on the target machine (most often: a third-party effect that isn't installed there), it's named explicitly in the result rather than dropped silently.

SELECT LAYER

Copies the currently selected layer as described above. If multiple layers are selected, only the first one becomes the source. Clears any chosen preset. Reports what it found, e.g. "Copied 'Title': 2 effects, transform anim (14 keyframes).", or tells you plainly if there's nothing to copy.

CLEAR

Empties the temporary clipboard.

Preset dropdown

Lists every .ffx file in the presets folder (see §17). Choosing one clears any layer copy. The × button next to it resets to no preset.

• PRESETS FOLDER: opens the folder in your file browser so you can drop new .ffx files in.
• RELOAD: re-reads the presets folder so newly-added presets show up in the list, sorted alphabetically.

Why presets are the answer for text animators & complex animation

KeyframeFX's Select Layer copy deliberately does not try to rebuild text animators, shape animation, or anything else complex: After Effects' scripting API won't let a script recreate the insides of a text animator without losing properties. An .ffx preset is AE's own native format and already holds everything AE can save, losslessly. So: build the animation, Animation → Save Animation Preset… in After Effects, drop the .ffx into the presets folder, and apply it from here: the panel only retimes what the preset laid down, it never tries to reconstruct it.

Preset Timing (only shown once a preset is chosen)

• AS SAVED (default): applies the preset exactly as it was saved. Same length on every target layer, however long they are.
• FIT LAYER: stretches or squashes the preset so it runs from the layer's in-point to its out-point. Drop the same preset on a 1-second clip and a 10-second one and each gets a version that fits.
  • Only the preset's own keyframes move. Anything already animated on the target layer stays exactly where it is: Fit Layer works out which keys the preset itself added and retimes only those.
  • The easing graph survives the retime: each keyframe keeps its influence, and its speed is divided by the same ratio the times were multiplied by, which is exactly what preserves a bezier's shape when the time scale changes. A preset that snaps hard out of its first keyframe still snaps the same way after being stretched to 4× the length.
  • Retimed keys land on whole frames wherever that doesn't push two of them onto the same frame.

Options (2-column toggle grid)

Effects/Transform/Audio are exhaustive: untick all three and PASTE does nothing at all (and tells you so).

PASTE

Applies whichever source is active onto every selected layer. Button text cycles PASTE → PROCESSING → DONE → back to PASTE. If a target property already carries an expression, the expression wins: the panel tells you it was skipped for that reason rather than silently failing.

UNDO LAST

Reverts the entire last paste run in one step (it drives AE's own Undo command: Ctrl/Cmd+Z does the same thing).

(tab icon tooltip: "Graph: bezier easing curves"): the custom cubic-bezier easing editor.

The curve plot

Drag either handle directly on the curve, or click the pencil next to the value readout to type exact numbers (format x1, y1, x2, y2). The readout itself (e.g. 0.33, 0.00, 0.67, 1.00) can be clicked to copy those values.

• X (time) is clamped 0–1. Y is allowed from −1 to 2, so bounce/ overshoot curves (like the built-in "Back" preset) are possible: the view automatically zooms out to keep an out-of-range handle in frame, and snaps back to the normal 0–1 view once both handles return inside it.
• Resize grip below the plot: drag to resize the graph area, double-click to reset to default height.

Toolbar icons

Fullscreen hides the app title, Quick Create and tab bar so the curve and preset list get the full panel: the icon becomes an "exit" glyph while active. It is not remembered between sessions, and switching to any other tab exits it automatically.

Quick-ease shortcuts

Three small buttons under the value readout jump straight to a common curve: Ease Out, Ease (in-out), Ease In.

APPLY EASING

Applies the current curve to the selected keyframes, or, if fewer than 2 keyframes are explicitly selected on a property, to every consecutive keyframe pair on that property. Only neighbouring selected keyframes form a pair: selecting keys 1 and 3 alone eases nothing. Both keyframes of every eased pair are switched to Bezier interpolation as part of applying it. One undo step.

VALUE GRAPH / SPEED GRAPH

Toggles the plot between the raw value curve and its speed (velocity) curve: useful for reading how fast the motion feels at each point, not just where it ends up.

Preset tabs

Four tabs, switched with a segmented control:

• BASE PRESETS: fixed, always available: Linear, Smooth, Ease In, Ease Out, In Out, Quad, Cubic, Expo, Sine, and Back (an overshoot curve).
• SPEED RAMP: fixed: Slow→Fast, Fast→Slow, Slow-Fast-Slow, Whip, Anticipate, Overshoot.
• MY PRESETS: your own saved curves, organised into named profiles so curves for different projects (or different people sharing the panel) don't pile into one list. Switch profiles from a dropdown; + makes a new profile, the pencil renames the current one, × removes it (at least one profile must always exist). Each preset can be renamed (double-click its label) or deleted (its own ×). Capped at 24 presets per profile: the oldest is dropped once you go over, and the status line names the one that went.
• IMPORTED: one or more named "packs" brought in from a file (see Import below), switched from their own dropdown. Imports never get merged into My Presets, so your own list can't be overwritten by someone else's file.

Import / Export

• IMPORT: brings in a .cgf file (Cepter EFX's own format) as a new pack under IMPORTED. Curve values may be a 4-number array or an "x1,y1,x2,y2" string: either works.
• EXPORT: saves whichever preset tab/profile/pack is currently shown as a .cgf file, to back up or share. Suggested filename matches the source's name; default save location is your Documents folder.

(tab icon tooltip: "SRT caption editor"): a segmented control at the top switches between two independent tools that share one caption format:

• AUTO CAPTIONS (beta): transcribes speech into caption layers, offline.
• SRT IMPORTER: imports/edits/creates captions from an existing .srt, or reads back captions built by Auto Captions.

10.1 Auto Captions

Runs OpenAI's Whisper models locally, through whisper.cpp. Your audio never leaves your computer: no account, no per-minute fee, works offline once the model is downloaded. (Everything here runs on your machine. No Python, no account, and nothing is uploaded.)

One-time setup

The first time you open Auto Captions (or whenever the engine, decoder, or every model is missing), you land on a setup screen instead of the transcribe screen:

• Speech engine: a DOWNLOAD button installs both the whisper.cpp engine and the shared audio decoder (FFmpeg) together, about 110 MB total. A CHECK FOR ENGINE UPDATE button tells you whether this version of Cepter EFX comes with a newer engine build than the one installed. It checks locally with no network call and never touches your models; when there is one, the setup button reads UPDATE and installs it.
• Model folder: shows where model files are kept, with BROWSE (to put them on a different drive: they're large) and DEFAULT (back to the folder inside the extension).
• Models: a scrollable list of 10 Whisper builds you can download individually (see the full table below), each with a live download progress bar (percentage, MB, elapsed time) and a CANCEL button. Deleting an installed model needs two clicks (the button becomes SURE? for a few seconds, then deletes).
• A BACK button appears once the engine, decoder and at least one model are all installed, so you can jump straight back to transcribing.

Whisper models are never silently "updated" once installed: a model is a fixed, permanent snapshot; improvements ship as a new named model (e.g. Large v3 → Large v3 Turbo), not a background update to one you already have.

Choosing what to transcribe

• DETECT SELECTED: reads the source file straight off whatever layer(s) you have selected in the timeline. Every qualifying selected layer becomes its own segment (sorted into timeline order and joined), so a chopped-up scene across several layers still transcribes as one continuous pass, with each piece's timing tracked back to its own position. Layers without a real source file (solids, precomps, text) are skipped automatically.
• BROWSE: pick any video/audio file from disk instead (whole file, starts at the beginning of the comp).

A hint line under these two buttons always states exactly what's currently loaded (which layers, how much audio, what portion of the timeline it covers).

Transcribe options

• Spoken language: 34 languages plus Auto detect (the default).
• Output: Original language (default, i.e. plain transcription) or Translate to English.
• Model: only shows models you've actually downloaded. Picking an English-only (.en) model automatically locks language to English and output to "Original language" (both dropdowns grey out): those builds can't translate and can't transcribe anything else.
• A rough time estimate appears once a model is picked, e.g. "1:24 of audio · roughly 0:12–0:37 on 7 threads": an estimate only, not a guarantee.

TRANSCRIBE

Renders the selected audio, then runs it through Whisper locally. After Effects is busy extracting the audio at the start of this, but nothing is polled afterward: the transcription itself runs fully in the background. A PROGRESS LOG panel shows live phase/progress, with COPY LOG (for sending a full diagnostic dump if something goes wrong) and CANCEL buttons. Large timeline selections are capped at 40 pieces in one pass: precompose first if you're over that.

The transcript editor

Once transcription finishes, you get an editable transcript with a live word count. Fix any misheard words here before building the layers: edits keep each word's original timing. (Editing works by re-splitting the text on whitespace and re-pairing it word-for-word against the original timings, so it's best for correcting individual words, not for casually adding/removing several words at once if precise timing matters.)

• Words per caption slider (1–20, default 4): the maximum words a single caption will ever show. It's a hard ceiling, not a target: a caption still breaks earlier than the limit at the end of a sentence or a natural pause, so one line never bleeds into the next.
• Max tail (seconds) slider (0.5–10.0, default 1.5): if there's silence right after a caption, how long it's allowed to keep showing before disappearing (never past where the next caption starts).
• GENERATE CAPTIONS builds bottom-centred text layers on the timeline, sized to the comp, coloured with the Captions tab's base colour (set in SRT Importer, see below). START OVER discards the transcript.

Generating captions leaves you on the Auto Captions pane: it does not jump you to SRT Importer automatically. To reword or recolour what was just built, switch to SRT Importer and press Sync From Timeline.

10.2 SRT Importer

An editable caption list you can build three ways: importing a .srt file, syncing captions Auto Captions already built, or just typing your own.

IMPORT .SRT FILE

Loads a standard .srt file's timings and text into the list below. Nothing touches your timeline until you press Create Captions.

Colouring captions

• Base colour swatch: sets the colour for every caption. Opens the shared colour picker.
• Highlight row: select (drag-highlight) text inside any caption below, then click a colour to recolour just that selection. Nine fixed one-click swatches (white, red, orange, yellow, green, cyan, blue, purple, pink) plus a custom-colour swatch (opens the full picker), a strip of recently-used custom highlight colours, and a reset swatch to clear a highlight back to the base colour.

The caption list

One row per caption: its time range, an editable text field you can type directly into, and a × to delete that row from the list (this doesn't touch anything already on the timeline: it only edits the in-panel list until you press Create Captions).

CREATE CAPTIONS

Builds bottom-centred AE text layers from the list: same placement/sizing rule as Auto Captions. Highlighted portions become their own colour via a text animator, so a single caption can show multiple colours at once.

SYNC FROM TIMELINE

Reads caption layers already on the timeline back into this editor, including anything you've since moved, retimed, retyped or deleted directly in After Effects. This reliably reads back layers Auto Captions built; captions created by SRT Importer's own Create Captions are not tagged for this same round-trip, so treat "sync" as primarily an Auto-Captions companion feature.

EXPORT

Saves the current caption list to a folder you choose, in any combination of five formats (tick as many as you want):

Files are written as captions.<ext> in the chosen folder.

(tab icon tooltip: "Voice Isolation: split dialogue from music & noise (beta)")

Splits a clip's audio into separate stems, offline, using the open-source demucs.cpp engine (a native C++ port of Meta's Demucs). Nothing is uploaded and there's no Python to install. (Everything runs on your machine. No Python, no account, and nothing is uploaded.)

One-time setup

• Separation engine: ships inside the extension: the CPU engine (demucs.cpp) and the GPU engine (ONNX Runtime with DirectML). LOCATE only appears if the bundled engine is missing, so you can point the panel at a compiled demucs.cpp.exe yourself.
• Audio decoder (FFmpeg): shared with Auto Captions and Dead Frames; installing it from any one of those three tabs satisfies all three (~100 MB, one time).
• Model folder / Models: same Browse/Default folder controls and per-model download-with-progress cards as Auto Captions, for the two separation models (see the model table below). A model shows as "installed" if it exists in either CPU or GPU format: switching between CPU/GPU never makes an already-downloaded model look missing; whichever format you're missing for your current engine choice downloads automatically the first time you press Separate.

Choosing a source clip

• DETECT SELECTED: reads the source off your selected timeline layer (single layer only). Uniquely among the panel's audio tools, this respects the layer's actual timeline edit: its trim (in/out point), start time, and stretch/speed are all applied to the extracted audio, so a sped-up, trimmed, or reversed (negative stretch) clip is separated exactly as it plays on your timeline, and the returned stems line up precisely under it.
• BROWSE, or drag and drop any audio/video file directly onto the source box (even from outside After Effects): both use the whole file, untrimmed.

What to separate

• ISOLATE VOICE (default): pulls just the vocal out, leaving music and background behind. Good for cleaning up dialogue, or prepping audio for Auto Captions.
• FULL STEMS: splits into vocals, drums, bass and "other" (a 6-stem model adds guitar and piano too), and you choose which to keep.

Both modes run the identical separation pass with the identical model: the mode only decides which stems are pre-ticked on in the results afterward.

ADVANCED (collapsed by default)

• Quality/model dropdown: choose between the two models (4-stem "Balanced", or 6-stem "Full band"). Defaults to the balanced 4-stem model.
• Processing: if your machine has a usable GPU, a live banner names it and a 3-way control lets you choose:
  • AUTO (default, recommended): GPU when available, otherwise CPU.
  • GPU: always use GPU acceleration (DirectML).
  • CPU: always use the CPU engine. On a machine with no usable GPU, this section simply states it's using the CPU and names it. GPU detection is a live, one-time probe each session: no hardware details are ever stored.

SEPARATE

Extracts the audio, then separates it locally. Big models on CPU can take several minutes: like everything else, it runs in the background. Your AE project is saved automatically first (if it's ever been saved to a file), as a safety step. A PROGRESS LOG with COPY LOG and CANCEL mirrors Auto Captions' working screen.

The stems mixer

Once separation finishes, every stem gets its own row:

• Play/pause: preview that stem (only one plays at a time); click anywhere on its waveform to seek.
• On/off toggle: include this stem in a combined import or not. Pre-ticked according to the mode you chose (only the vocal stem for Isolate Voice; every stem for Full Stems).
• Volume: a slider (0–200%) and a typeable dB box (−60 to +6), live-linked to each other and to the currently-playing preview.
• Import (single stem): adds just that stem to the timeline at its set volume.
• Export (single stem): saves just that stem as a .wav.

Bottom actions:

• IMPORT ALL: every stem becomes its own separate audio layer on the timeline.
• EXPORT ALL: every stem saved as a .wav into a folder you pick.
• COMBINE & IMPORT: mixes every toggled-on stem at its own set volume into a single audio layer, imported under your clip.
• START OVER: discards the results.
• COPY LOG: available throughout this tab (not only while a job is running) for sending a full diagnostic dump if something needs debugging.

(tab icon tooltip: "Beat Markers: find the beat in your audio and mark it")

Points at the layers carrying your music, finds the hits, estimates the tempo, and drops markers on the composition, a layer, or a fresh null.

Picking the source

USE SELECTED LAYERS: reads whichever selected timeline layers actually carry audio (several selected pieces are joined in timeline order, so a chopped-up song still works as one continuous track). Anything selected that has no audio is silently skipped.

The waveform

Once analysed, a waveform appears with:

• Click empty space to add a beat immediately at that point.
• Click an existing beat to remove it.
• Drag a beat to nudge its exact position.
• Mouse wheel over the waveform zooms in/out, centred on the cursor.
• +/− buttons, a pan slider, and a fit-to-view button for navigating a longer track.

Nothing reaches your composition until you press Place Markers: the waveform is a safe place to review and fix the detected beats first. Changing Sensitivity, the Advanced settings, or the BPM re-runs detection and replaces any hand edits you made: do your manual click/drag fixes last, after you're happy with the detected settings.

Sensitivity

A single slider up front (5–95, default 55): how much of a jump in the music counts as a hit. Turn it up to catch more, down to keep only the obvious ones.

ADVANCED drawer

Collapsed by default, already set to what works best for most music. Its label changes to "ADVANCED - EDITED" the moment anything inside no longer matches the defaults, so a setting changed weeks ago is never invisible: one button (wand icon, bottom of the drawer) resets everything inside back to the defaults in one click.

• What to listen for: FULL MIX (default, best for sparse/acoustic music), KICK (only the low end: kicks/bass, right for most electronic music), or SNARE/HATS (only the top end). Changing this one control does not auto-re-analyse: press Analyse again after changing it.
• Min gap: the shortest allowed gap between two markers, in milliseconds (30–600, default 120). Raise it if one drum hit is being marked twice.
• How to place the beats:
  • TRACKED (default): follows a steady pulse across the whole track rather than trusting each hit in isolation, then nudges every beat onto the nearest real onset. Won't drop a beat in a quiet bar, and won't double-mark a busy fill.
  • HITS: every detected hit, exactly where detected, no assumption of a steady tempo. Right for live/rubato playing.
  • EVEN GRID: ignores the actual hits and lays down a perfectly even beat at the estimated tempo. Only sensible for programmed music with a fixed BPM.
• Thinning: EVERY beat (default), or every 1/2, 1/4 (roughly one marker a bar in 4/4), or 1/8 beat.
• BPM: a text field defaulting to auto; type a number to force it (useful when a track reads as exactly half or double the real tempo), a TAP button (tap along 4+ times to set it by ear), and an AUTO button to go back to the detected value.
• Where the markers go:
  • COMP (default): composition markers on the comp's own marker track.
  • LAYER: layer markers on whichever layer is selected at the moment you press Place Markers.
  • NEW NULL: a fresh null layer called "Beats", spanning the whole comp, carrying the markers: keeps them out of the way and easy to delete as one layer.
• Toggle cards:
  • SNAP TO FRAMES (default on): rounds every marker to the nearest whole frame.
  • NUMBER THEM (default off): numbers each marker, marking every 4th one as a bar.
  • REPLACE EXISTING (default on): wipes any markers already on the chosen target before placing new ones; turn off to add a second pass on top of what's there.
  • CLEAR MARKERS: removes every marker from the currently-chosen target without placing anything new.

ANALYSE

Extracts the audio, then finds the beats: nothing is written to your comp yet. With the shared audio decoder (FFmpeg) installed, this uses an accurate spectral-analysis pass; without FFmpeg it silently falls back to After Effects' own "Convert Audio to Keyframes," which is noticeably coarser (one value per frame): the panel tells you when this fallback is active and suggests installing FFmpeg from the Captions tab. Caps at 40 audio pieces in one pass.

PLACE MARKERS

Writes the currently-found (and any hand-adjusted) beats into your composition as markers, on whichever target you chose above. One undo step.

(tab icon tooltip: "Precomp: full precompose with names, safety & render queue")

Note: the tool that precomposes selected layers with automatic naming ("PRECOMP SEP") actually lives on the Tools tab: see §7. This tab's own content is the QueueForge batch render-queue tool below, plus a currently-disabled Count Layers button.

QueueForge: batch render queueing

Stages multiple comps into After Effects' real Render Queue in one pass: it does not start rendering itself; you still review the queue and press Render yourself.

1. Press the small refresh icon to load the comp list and available Output Module format templates. (This is deliberate and manual: reading your saved templates briefly pops AE's own Render Queue panel forward, so the panel waits for you to ask for it rather than doing it the moment you open the tab.)
2. Format dropdown: pick an Output Module template. (Save these once in AE's own Render Queue first, then they appear here.)
3. CHOOSE FOLDER: pick where the rendered files will be written.
4. Output name field: sets the output filename used when these are added to the queue. It does not rename the comps themselves. Leave it blank to use each comp's own name.
5. Comp checklist: every comp in your project, tick the ones you want to queue. Select all / Clear all toggles the whole list at once.
6. ADD TO RENDER QUEUE: adds every ticked comp with the chosen format and destination folder. Grey/disabled until you've picked at least one comp, a format, and a folder. If two queued comps would produce the same output filename, a _1, _2… suffix is added automatically so nothing overwrites anything else in the same batch.

One undo step covers the whole batch add.

(tab icon tooltip: "Dead Frames strip anime's held/duplicate frames, keep the moving ones")

Anime is drawn "on twos and threes": each drawing held for two or three frames, with long runs of identical frames between the ones that actually move. Dead Frames finds the drawings that really move and rebuilds the clip as a Time Remap that plays only those, back to back: the tedious first half of a Twixtor-style pass, done in one press.

Select one real video footage layer (not a precomp, solid, or text layer), then press Remove Dead Frames below: there's no separate detect step; it reads whatever layer is selected in your timeline automatically as part of that one press.

Sensitivity

• SMALL: most sensitive; keeps even tiny motion (drifting particles, subtle shading). Use when preserving every scrap of movement matters more than the cleanest strip.
• BALANCED (default, recommended): a denoise pass strips compression grain/shimmer before comparing frames, so real drawings are kept and noise artefacts aren't mistaken for motion. Best all-round setting for typical anime.
• STRONG: only counts clear movement; the cleanest strip on grainy or low-bitrate footage.

Quality

• FAST (default): smaller analysis size, lighter denoise. Plenty for clean clips.
• ACCURATE: a stronger denoise pass (not a bigger frame; a larger analysis size was actually measured to hurt accuracy): much better at ignoring compression grain on noisy/low-bitrate footage.

Analyse an Area

Off by default. Turn it on to restrict analysis to part of the frame: switching it on reads your timeline selection and loads the frame under your playhead at the clip's real proportions automatically, then drag a box over the character or region that matters: motion elsewhere in the frame is ignored.

Close Gaps

A general keyframe tool that lives here because it pairs naturally with condensing a clip, but works on any selected layers or keyframed properties (not only Dead-Frames-processed clips): packs spread-out keyframes onto consecutive frames, one right after the previous.

REMOVE DEAD FRAMES

The one button that runs the whole tool: detects the selected layer, analyses it, and rebuilds it as a Time Remap that plays only the kept frames, trimming the clip to its new, condensed length, all in a single press. One undo step.

Touch-up: REMOVE SELECTED FRAMES

No detector is perfect. If a held frame slips through, select its Time Remap keyframe on the already-cleaned layer and press this: it deletes the selected keyframe(s) and re-packs the rest so the clip stays gap-free. One undo step.

Requirements

Needs the shared audio/video decoder (FFmpeg): a one-time ~100 MB download, shared with Auto Captions and Voice Isolation, and a real footage layer, never a precomp, solid, or text layer.

(tab icon tooltip: "About: panel settings, version, socials"): this tab always opens on the ABOUT sub-pane, whichever one you last had open, via a segmented control: SETTINGS · UPDATES · ABOUT.

Settings

Precomp group

• Safety Engine (default on): duplicates your comp into a comp named "Safe Master" before Precomp Sep changes anything.

Performance group

• Auto Purge (default off): watches how large After Effects' disk cache has grown and clears it once it passes a size you set. Worth knowing exactly what this can measure: After Effects gives scripts no way to read its RAM usage, so the number being watched is the disk cache folder size on your drive: the only figure that can actually be measured; the purge itself still clears both memory and disk. When turned on:
  • Clear over (N MB) slider: 200–20,000 MB, default 4000.
  • Check every (N min) slider: 1–60 minutes, default 10. Checking means asking After Effects to briefly walk the cache folder, so longer gaps are kinder than short ones.
  • CHECK NOW: measures immediately, reporting the size, without purging (purging only ever happens automatically once the size is over your limit).
  • CACHE FOLDER: point at the cache folder yourself, only needed if After Effects won't report where it is (you can find the real path in After Effects under Edit → Preferences → Media & Disk Cache).
  • The very first automatic check after turning this on is deliberately delayed 90 seconds, since right after launch is the worst moment to hand AE a folder-walk.
• Static Mode (No Animations) (default off): turns off every animation, glow, transition and background effect across the whole panel so it renders completely static. Turn this on if the panel feels sluggish on a lower-end machine: besides the visual cutback, it also stops the star background from spawning particles and the hover-glow effect from tracking your cursor, cutting real work, not just visuals. Progress bars stay animated regardless (they convey real information).

Theme colour group

• A colour swatch (opens the full picker) and a hex field: sets the accent colour used across the entire panel. Ten one-click presets: Purple (default), Blue, Cyan, Green, Lime, Amber, Orange, Red, Pink, Magenta. RESET returns to the default purple. Picking a full custom colour sets its exact hue, saturation and lightness, not just a re-tint of purple.

Appearance group

• Interface Font: a 3-way choice: System (Segoe UI on most Windows machines, nothing extra loaded), Inter (default): sharper at the small sizes this panel uses, or Poppins: a rounder, brand-forward look (letters run a little wider, so the tightest labels sit close to their edges). Both custom typefaces ship inside the extension: nothing is ever downloaded, and this works fully offline.
• Ambient Stars (default on): the floating background particles. When on:
  • Moving Particles (default off): makes the stars drift instead of only twinkling in place.
  • Rainbow Particles (default off): gives stars randomised vibrant colours instead of the theme colour.
  • Star Count slider (1–50, default 5): density of the background stars.

Updates

Shows the version and release date you're running, and the release notes for this build only: entirely local, computed from constants baked into the build. The panel never checks a server for a newer version: this section is not a live "check for updates," just a local readout of what shipped in the copy you have.

About

Credits, and two buttons: ALL LINKS (Instagram, TikTok, YouTube, one page) and DISCORD (bug reports, feature requests, or just to say hi).

One HSB/RGB/hex picker is used everywhere a colour needs picking: the Quick Create solid swatch, the Captions tab's base/highlight swatches, and the About tab's theme swatch. Its title changes to match what you're picking (e.g. "SOLID COLOR", "THEME COLOUR").

• A saturation/brightness square and a hue strip, both draggable.
• Numeric H/S/B and R/G/B fields, plus a hex field: type or paste any of them and the rest update live.
• Current vs New preview swatches so you can compare before committing.
• Recent Colours: the 8 most-recently-used colours, shared across every picker use in the panel (picking a colour anywhere adds it here; each recent swatch has its own small remove button).
• Cancel / OK at the bottom; clicking outside the picker also cancels.

Drop .ffx animation presets into the extension's presets folder. In the panel: KeyframeFX tab → PRESETS FOLDER jumps straight there: drop files in, then press RELOAD. Your presets then appear in the preset dropdown and are applied when you press PASTE.

To make one: select the effects/keyframes you want to save on a layer in After Effects, then Animation → Save Animation Preset…, and save the .ffx into that same folder.

Every toggle, slider, colour choice, theme, and last-used value anywhere in the panel is written to your browser-local storage the instant it changes, and silently restored the next time you open the panel: nothing needs saving by hand, and nothing is ever synced anywhere else. Updating to a newer version of Cepter EFX doesn't touch any of it.

All AI processing (Auto Captions' transcription, Voice Isolation's stem separation, Beat Markers' analysis, Dead Frames' frame analysis) happens entirely on your own machine. Nothing about your footage, audio, or project is ever uploaded anywhere. The only network activity the panel ever initiates is the downloads you trigger yourself from a setup screen (engines, models, FFmpeg).

• A tool's setup screen won't download anything. Check your network: common causes surfaced in plain language include DNS not resolving (a VPN, proxy, or network filter is the usual culprit) or a certificate that can't be verified (often antivirus HTTPS inspection, or a wrong system clock). A cancelled or failed download always cleans up its partial file, so it can never look "installed" when it isn't.
• Something in Auto Captions or Voice Isolation went wrong mid-job. Press COPY LOG (present on both tabs' working screens, and persistently in Voice Isolation) and send that; it's a full diagnostic dump: version, timestamps, engine/decoder paths and whether they're present, installed models, the input file, and the complete run log.
• A job seems stuck. Both features have an internal watchdog that gives up and reports "The job stopped responding" if a process goes quiet too long: it will tell you rather than hang forever; check for a leftover command window if this happens and close it before retrying.
• The panel feels sluggish on your machine. Turn on Static Mode in About → Settings, see §15.
• After Effects itself gets slow. Use PURGE on the Tools tab, or set up Auto Purge so it happens automatically.
• General help, bug reports, feature requests: the Discord link and "All Links" button on the About tab, or linktr.ee/intercepth.

Ten models total: six multilingual, four English-only builds of the smaller sizes (Large has no English-only version; OpenAI never trained one). An English-only model is the same size and speed as its multilingual counterpart, just a little more accurate on English since none of its capacity is spent on other languages, but it can't translate and can't transcribe any other language.

Both are local builds of Meta's Demucs (htdemucs architecture), available in a CPU format and, if you use GPU acceleration, a separate GPU-optimised format: either counts as "installed" once you've downloaded it in one format.