agentleFS
alisadiq-aialisadiq-ai/favstash-ai-video-editing-skills/skills/3-styles-on-camera/adaptive-glass/SKILL.md

adaptive-glass

Skill75 starsChanged 40 days agoApache-2.0View on GitHub

Style for edit-video: the real speaker with moving proof in footage-matched glass panels, centered captions and event-matched sound. Use for talking-head product, tool and explainer reels with real screen proof, or when the creator asks for the glass look.

The file

We split this skill into the rules it gives an agent

8 rules. About 0 are common, 5 are distinctive, and 3 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈5

    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
  1. Distinctive:

    Start proof-heavy reels with the half split when the proof stays readable.

    line 27
  2. Distinctive:

    Use the actual camera plate, keep its aspect ratio and set the crop from real frames of each take.

    line 36
  3. Distinctive:

    Keep a glass surface at full opacity while it enters and move it instead, about 18 px over .42 s with an ease-out.

    line 63
  4. Very specific:

    Use edit-video's sound pack (by id) and the gains in its manifest as starting points:

    line 101
  5. Distinctive:

    Keep speech dominant and make demonstrated clicks and reveals audible.

    line 115
  6. Very specific:

    Write a scene plan. Start from the example: frame ranges at 30 fps, layout (face, split-half, split-third, full-proof), type (none, …

    line 123
  7. Very specific:

    Check, inspect key frames, then render with the workspace's HyperFrames (prepared by edit-video's init-workspace.mjs --install):

    line 137
  8. Distinctive:

    Treat the output as a starter, not an art director.

    line 149

≈ “Common” is an estimate: at least 1 in 5 of the 372,429 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.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Adaptive glass

A style on top of edit-video, which owns the cut, captions, sound and review. The finish for talking-head product, tool and explainer reels: the real speaker, moving proof, a palette taken from the footage, centered captions, selective modern glass and sound that follows visible events. Use it when the creator chooses it or has no established look. Their own fonts, colors and references take priority. Full camera, split screen and full proof are layouts within the style, not separate styles.

Skip it for material that needs no panels: a personal reflection, a joke over footage or a long screen demo that needs the whole canvas.

Layouts follow the material

LayoutUseStarting geometry at 1080×1920
Full cameraA direct statement, reaction or turn with no useful assetCamera fills the canvas; captions in the headroom
Half splitProof or a diagram reads clearly above the speakerContent above y=960; camera pane y=960–1920
Third splitProof needs more height and the face stays completeCamera pane from about y=1120
Full proofClose UI detail, dense text, a result or an animationCamera hidden; enlarge or reframe the useful detail

Start proof-heavy reels with the half split when the proof stays readable. Full screen has to earn its area; small proof over a large empty background means the face should come back. Screen recordings fill their pane edge to edge (media-bleed), without a nested card or heading. Change layout on spoken beats or reveals. Open on the hook's subject: frame 1 should already show the product or proof, not an empty panel.

Face treatment

Use the actual camera plate, keep its aspect ratio and set the crop from real frames of each take. The split pane blends through spare headroom: transparent at the top edge, about 12% opaque at 65 px, 55% at 160 px and solid at 270 px, with a 75 px side feather. Keep eyes and mouth sharp, keep the whole face in the pane and check the blurred fill for a second ghost face. Full and split camera layers are never visible together outside a deliberate transition. A tension hook can use a fast, obvious push; compare first and peak frames, because a speaker leaning back can cancel a digital zoom.

Palette from the footage

Inspect the shirt, room, skin and proof separately. A dark shirt suits a lighter related field; a light shirt may need a deeper one. The starting palettes are soft-violet, deep-violet and charcoal; choose one per reel with palette and override individual tokens with a paletteFile. Keep one color family, use dark heading ink on a light field and light text on deep glass, and use red or green only when they mean something. Polish the camera plate from the take (exposure, white balance, natural skin) before judging colors.

Material

  • glassMaterial: "optical" for main panels, "soft" for supporting information; glassTone: "dark" or "light". Set them for the whole plan or per scene.
  • One quiet 1 px perimeter and 28–36 px corners. No stacked corner strokes. Nested media corners are concentric: inner radius ≈ outer radius minus the inset.
  • Keep a glass surface at full opacity while it enters and move it instead, about 18 px over .42 s with an ease-out. Fading an ancestor changes what the backdrop samples and makes the blur pop. Fade text separately if needed.
  • At most one low sweep after an optical panel settles, and only when the shot is long enough. Never a perpetual shimmer. Keep the blur stable.
  • The CSS is optical styling, not refraction. For liquid-glass lensing, morphs or a named effect, search the HyperFrames registry (hyperframes catalog), inspect the item and render-test it in the project. If true fluid merging does not render reliably, use an honest shared-shape expansion and say so.
  • Proof footage stays unobscured. Full proof is content, not a frosted panel.

Motion

Carry one surface through connected states: a chip becomes a panel, a panel slides into a new position, related shapes join or split when the explanation calls for it. Tie each move to a spoken beat or visible action. Small arrivals take about .3–.5 s and larger shared-shape transitions .45–.8 s, followed by a readable hold. Use continuous velocity and a restrained settle; avoid a universal bounce, robotic linear slides and repeated identical pop-ins. Words stay stable while their surface moves. Every rendered state comes from timeline time.

Type and captions

Helvetica Neue Bold is the default, with Helvetica and Arial as fallbacks. Use an installed font and confirm the renderer used it; never commit system font binaries. Captions are usually one or two lines at about 54 px on a 1080-wide export; panel headings 52–76 px with few words; source labels 24–30 px. Headings, eyebrows and footers are optional. Alongside captions, allow at most one short extra emphasis.

Starting caption anchors (fraction of height): full camera .235, half split .55, third split .595, full proof .60–.615. Two-line captions overflow easily, so measure the rendered box, backing and shadow against edit-video's safe area. The scaffold writes these anchors and a starting caption style to captions.json for the captioning tool.

Sound map

Use edit-video's sound pack (by id) and the gains in its manifest as starting points:

Visible eventSoundTiming
Frame 1 of the hookhook-impactOn the opening frame
Large layout or camera changecamera-shutterAbout 4 frames before the cut
Small framing or take changecamera-shutter-lightAbout 2 frames before
Panel or element arrivalui-popOn the arrival frame
Visible click or selectionui-clickOn the click
Motivated zoom, crop or panzoom-swishIts rise follows the move
Build into a revealmetallic-riserTrim or stretch it to resolve on the reveal
Earned completionclean-ding.wavSparingly

Keep speech dominant and make demonstrated clicks and reveals audible. Put each cue family on its own track so overlapping cues are not replaced, merge cues fewer than about 6 frames apart and keep ordinary caption changes silent.

Build the visual layer

  1. Prepare the camera plate and proof clips in the run's assets/: trim, and crop from the original high-resolution source rather than enlarging a small copy.
  2. Write a scene plan. Start from the example: frame ranges at 30 fps, layout (face, split-half, split-third, full-proof), type (none, media-bleed, media-card, media-wide, proof, diagram), media ranges, optional label, title, footer or captionY.
  3. Generate a new composition:

    node <skill>/scripts/scaffold-glass.mjs <run>/work/scenes.json --output <run>/work/glass-v01

    It validates the whole plan before writing, refuses to overwrite a revision, copies media by hash and copies GSAP from the workspace runtime prepared by init-workspace.mjs --install.

  4. Check, inspect key frames, then render with the workspace's HyperFrames (prepared by edit-video's init-workspace.mjs --install):

    .favstash-studio/runtime/node_modules/.bin/hyperframes check <run>/work/glass-v01
    .favstash-studio/runtime/node_modules/.bin/hyperframes render <run>/work/glass-v01 --fps 30 --quality high --output <run>/previews/glass-v01.mp4

    The composition is silent: assemble dialogue, captions and sound in the editor or with FFmpeg, keeping them editable. The checker's duplicate_media_discovery_risk warning is expected, because the full-camera, split and blurred-fill layers share one camera plate.

  5. Treat the output as a starter, not an art director. Adjust face placement and proof crops from the footage, and add purpose-built motion where proof is missing.

The material CSS is an original implementation informed by the six-layer optical model in liquid-glass-css; no third-party code is included.

Download now (ZIP)

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
What did you use it for? e.g. Reviewing pull requests in a Rails app

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.