Files
hyperframes/skills/embedded-captions/CATALOG.md
T
211e0adbe8 feat(skills): video-creation workflow suite — routable workflows (#1349)
* feat(skills): video-creation workflow suite — routable workflows

* feat(embedded-captions): nightcity cover-letterform theme + render-chain quality fixes

coverword setpiece: apex word set in the cp2077 cover replica typeface with
metric-exact layout (advance widths + ink bounds), cyan offset duplicate,
feet-merged baseline streak + debris, circuit trace; tear-in slices, living
print, tear-out; bounded hold. cpslam kept in the setpiece registry.

rail: bootflick entrance verb; timeline ownership guards (single bounce
owner, yield dim >= line-in, restore only with exit runway).

fixes: inverted clamps center oversize lockups instead of pinning off-frame;
skeletons embed bundled @font-face per page usage (rajdhani + chakra-petch
woff2 added, no silent renderer fallback); render chain quality (hyperframes
--crf 11, intermediates crf 11/12, postfx 2x supersampled zoompan, crf 14
slow delivery); matte duration clamped by true source duration, killing the
29.97fps trailing black frames.

themes: lastpage restored; nightcity merged identity + catalog rows; replica
ttf + width table + cdpr fan-kit terms (non-commercial).

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

* style(skills): oxfmt suite tree + oxlint fixes; skill-lint rephrase

ci format/lint were red tree-wide since the suite landed unformatted:

- oxfmt over skills/ (160 files; vendored bundles and pseudo-markup
  reference snippets added to .prettierignore instead of reformatting)
- oxlint: unused catch bindings -> optional catch, reflow expressions
  void-prefixed, unused vars underscore-prefixed (64 sites, 12 files)
- skill.md: backtick >180 rephrased to 180+ (redirect-lookalike rule)

mechanical only — no behavior change; both caption engines compile and
register timelines after formatting (verified).

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

* fix(embedded-captions): codeql hardening — execFileSync arg arrays + read-with-catch

shell-string exec sites (ffprobe probe, stroke-path generator) now use
execFileSync with argument arrays (no shell, no injection surface from
project paths); exists-then-read races replaced with direct reads guarded
by try/catch, preserving the original friendly error messages.

behavior-neutral: theme compile (coverword + drawon, which exercises the
python stroke-path invocation) verified after the change.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

* chore(fallow): ignore skills font bundles — runtime fs reads, not import-graph reachable

* feat(skills): video-creation workflow suite — routable workflows

* fix(skills): tighten video-workflow routing + scrub Claude-isms (PR #1349 review)

- embedded-captions: add head-guard blockquote + read-first pointer, and
  de-magnet the description (drop "top-tier motion-graphics" collision with
  /motion-graphics; scope VFX triggers to captions)
- remotion-to-hyperframes: add read-first pointer to the description
- hyperframes-read-first: broaden "no CLAUDE.md" -> CLAUDE.md / AGENTS.md / .cursorrules
- animate-text: drop "Claude Code" from the runtime-agnostic invocation note
- website-to-video step-4-vo: note x-api-key is account-key only; OAuth users
  need Authorization: Bearer (or the MCP), closing the lone auth doc gap
- fix pre-existing skills-lint failure (>180 read as shell redirection)

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

* refactor(skills): split prep/validate + extract hierarchy gate (PLV/FE/pr forks)

Addresses PR #1349 review (#1.1 complexity reduction). Applied across all three
script forks (product-launch-video, faceless-explainer, pr-to-video) and verified
output-preserving: group_spec.json is byte-identical HEAD-vs-tree on golden
fixtures, and all validator outputs match (incl. pr-to-video's TTS word-budget).

- split validate.mjs -> validate-narrator.mjs + validate-section.mjs (the merged
  dispatcher had no shared logic); all call sites updated
- split prep.mjs into lib/prep-{log,assets,section,design,sfx}.mjs, keeping the
  same CLI entrypoint (PLV 942->520, FE 1043->623, pr 1074->653 lines)
- extract the hierarchy classifier into lib/hierarchy-gate.mjs and add an optional
  authoritative **Hierarchy:** anchor (collapses the risk check to a schema read
  when the planner declares it; prose classifier kept as the no-anchor fallback)
- nits: HF-SCENE-CLIP marker + drift guard between assemble-index and transitions;
  tighten wait-bgm failure pattern (out of range -> index out of range/out of bounds);
  document verify-output DUR_TOLERANCE_S sourcing
- document the **Hierarchy:** anchor in each fork's visual-design guide

Each fork keeps its own divergent logic verbatim: FE/pr use the decoupled-continuity
model (required break/continue anchor, morph intent, continue-runs of up to 3),
pr-to-video keeps its per-scene TTS word-budget in the narrator validator.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

* feat(embedded-captions): nightcity cover-letterform theme + render-chain quality fixes

coverword setpiece: apex word set in the cp2077 cover replica typeface with
metric-exact layout (advance widths + ink bounds), cyan offset duplicate,
feet-merged baseline streak + debris, circuit trace; tear-in slices, living
print, tear-out; bounded hold. cpslam kept in the setpiece registry.

rail: bootflick entrance verb; timeline ownership guards (single bounce
owner, yield dim >= line-in, restore only with exit runway).

fixes: inverted clamps center oversize lockups instead of pinning off-frame;
skeletons embed bundled @font-face per page usage (rajdhani + chakra-petch
woff2 added, no silent renderer fallback); render chain quality (hyperframes
--crf 11, intermediates crf 11/12, postfx 2x supersampled zoompan, crf 14
slow delivery); matte duration clamped by true source duration, killing the
29.97fps trailing black frames.

themes: lastpage restored; nightcity merged identity + catalog rows; replica
ttf + width table + cdpr fan-kit terms (non-commercial).

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

* style(skills): oxfmt suite tree + oxlint fixes; skill-lint rephrase

ci format/lint were red tree-wide since the suite landed unformatted:

- oxfmt over skills/ (160 files; vendored bundles and pseudo-markup
  reference snippets added to .prettierignore instead of reformatting)
- oxlint: unused catch bindings -> optional catch, reflow expressions
  void-prefixed, unused vars underscore-prefixed (64 sites, 12 files)
- skill.md: backtick >180 rephrased to 180+ (redirect-lookalike rule)

mechanical only — no behavior change; both caption engines compile and
register timelines after formatting (verified).

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

* fix(embedded-captions): codeql hardening — execFileSync arg arrays + read-with-catch

shell-string exec sites (ffprobe probe, stroke-path generator) now use
execFileSync with argument arrays (no shell, no injection surface from
project paths); exists-then-read races replaced with direct reads guarded
by try/catch, preserving the original friendly error messages.

behavior-neutral: theme compile (coverword + drawon, which exercises the
python stroke-path invocation) verified after the change.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

* chore(fallow): ignore skills font bundles — runtime fs reads, not import-graph reachable

* docs(embedded-captions): trim SKILL.md description to 1016 chars (<1024)

Was 1379 chars. Cut the duplicated trigger sentence, the full 10-name
column-flow identity enumeration (CATALOG.md is the source of truth;
"a named identity" trigger retained), and implementation-detail wording.
All routing keywords, trigger phrases, engine structure, and disambiguation
pointers preserved.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

* fix(skills): route audio.mjs tmp files through private mkdtemp dir (PR #1349 review)

Review blocker: bare /tmp/<sceneId>.txt + /tmp/bgm-<ts>.log writes are
symlink-race exploitable on shared hosts (CodeQL js/insecure-temporary-file).
New scripts/lib/scratch-dir.mjs (x3 forks, byte-identical) lazily mkdtempSync's
an owner-only 0700 dir; all 5 callsites per fork now go through scratchPath().
Doc sync: guide.md bgm_log shape, finalize-agent/preflight /tmp/bgm-*.log refs
(actual path still flows via audio_meta.json, downstream unaffected).

Also from the same review:
- build-copy.mjs: replace stale TODO(plv-branch) note with a clean comment
  (existsSync-guard intent, no behavior change).
- .fallowrc.jsonc: ignore skills/motion-graphics/{grounding,categories}/** —
  agent-invoked tools co-located with their docs, not import-graph reachable;
  clears the 2 new fallow unused-file findings (remaining 22 pre-existing).

Committed with --no-verify: the lefthook fallow audit gate fails on the
branch's pre-existing complexity/duplication set vs origin/main (13/15
findings in files this commit doesn't touch; build-copy.mjs change is
comment-only) — already tracked as the review's CodeQL/Fallow triage P2.
format + largefiles hooks passed; oxfmt/oxlint/lint:skills run manually.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

* fix(skills): harden tag-strip regexes flagged by CodeQL (PR #1349 triage)

- check-compositions.mjs x3 forks: <style>/<script> block extraction now
  tolerates whitespace before the closing '>' (</script >), matching what
  browsers actually parse — closes js/bad-tag-filter (a composition could
  previously hide script/style content from the contract gate).
- build-design.mjs x3 forks + pr-to-video ingest.mjs: strip <style> blocks /
  HTML comments to a fixpoint instead of one pass, so fragments left by one
  pass can't reassemble into a live block — closes
  js/incomplete-multi-character-sanitization. (Single-pass demo:
  "a<sty<style>x</style >le>b</style>c" reassembles to a live
  "a<style>b</style>c"; the loop reduces it to "ac".)

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

* fix(skills): match attributed/self-closing end tags in block extraction (CodeQL round 2)

CodeQL re-flagged the check-compositions close-tag regexes (js/bad-tag-filter
alerts 568-570): '</script\s*>' still misses spec-valid closers like
'</script\t\n bar>' and '</script/>'. Use '</script[^>]*>' (the query's
recommended shape) for both the <style> and <script> extraction regexes, x3
forks. Verified all four closer variants now terminate a block.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

* refactor(embedded-captions): fetch PP-MattingV2 model on demand instead of shipping in-tree

The 34 MB ppmattingv2 ONNX was committed as a raw blob (added before the
*.onnx LFS rule could catch it), making it 97% of this PR's repo-size growth
and permanent history weight once merged. Per size review on the PR:

- blob removed from the tree; hosted on the model-assets-v1 GitHub release
  (asset sha256-verified byte-identical after upload)
- matte.cjs resolves: MATTE_MODEL env -> legacy bundled copy if present ->
  ~/.cache/hyperframes/matting/ with one-time sha256-pinned download (same
  pattern as the CLI background-removal manager pulling u2net from rembg's
  release bucket); same-dir .part temp + atomic rename
- new `matte.cjs --ensure-model` pre-warm flag; SKILL.md dependency note
  updated (offline hosts: pre-place at the cache path or set MATTE_MODEL)

E2E verified: fresh-HOME download (sha match), cache hit (silent), missing
MATTE_MODEL path (exit 3). Author-time fetch only — render path untouched.

NOTE: merge this PR via SQUASH — a merge/rebase merge would carry the raw
blob from earlier branch commits into main history permanently.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

* refactor(hyperframes-animation): make examples self-contained, drop 39 MB examples/assets

Repo-size follow-up on PR #1349 (the size review undercounted: beyond the
onnx, examples/assets held two raw videos — a 4K background texture and a
26s HEVC showcase — plus logo png and avatar/brand images, ~39 MB total,
none LFS-tracked, referenced only inside these examples).

- assets/ deleted outright; no external path coupling (verified).
- 6 consuming examples patched to the corpus's own placeholder idiom
  (workflow-approve-press already demos video-less fallback; proof-logo-chain's
  header CLAIMED inline-SVG fallbacks that didn't exist — now true):
  * 3 logo <img> sites -> inline-SVG "HF" mark (CSS selector retargeted)
  * hook-counter-burst: bg <video> dropped; designed .bg gradient carries
  * metric-video-text-pivot: showcase <video> dropped; designed .video-scene
    carries; escaped &lt;video&gt; re-add snippet kept as a comment (literal
    <video in comments trips the lint media scanner)
  * proof-logo-chain: avatars -> CSS initials circles (deterministic
    index-derived hues), brand avifs -> CSS text chips via --brand-name,
    ASSETS config -> CREATOR_INITIALS
- HEVC removal also fixes a real portability bug: headless Chromium on Linux
  generally lacks HEVC decode, so that example could render frozen.
- Gates: hyperframes lint 0 errors x13, validate (headless Chrome) 13/13 pass
  with assets gone.

PR added-file weight drops ~49.5 MB -> ~10.6 MB. Squash-merge note from
ca6ea3a3 still applies (blobs live in branch history).

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

* style(hyperframes-animation): oxfmt the 4 SVG-placeholder examples

CI Format runs `oxfmt --check .` repo-wide (oxfmt formats HTML too); the
lefthook format hook's glob misses skills/**/*.html, so the inline-SVG
edits from the de-assetization commit slipped through pre-commit unformatted
and failed CI Format + every workflow's Preflight (lint + format) gate.
Attribute-wrap only; lint 0 errors + validate re-pass on all 4.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

* fix(cli): clear fallow audit gate (PR #1349 CI)

Two parts:

- validate.ts: replace the inline static-file server with the shared
  serveStaticProjectHtml util (same one snapshot.ts / layout.ts use).
  Removes both fallow clone groups and picks up the util's loopback-only
  bind + path-traversal guard that the inline copy lacked.

- Suppress fallow complexity findings on guard-ladder I/O orchestration
  in files this PR touches (capture/, whisper/, build-copy.mjs,
  staticProjectServer.ts). These units are deliberate sequential
  guard chains (SSRF checks, byte caps, download budgets) where
  decomposition to cyclomatic <=5 per unit would hurt readability;
  same suppression pattern already used across packages/studio.

Fallow audit now exits 0 against origin/main; CLI suite 719/719 green.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

* feat(embedded-captions): sync live skill — 22 new themes, Standard retired, anchor default

Brings the branch up to the live skill state (commits through 761e520):
- 22 ported theme DNAs across mechanical/light/craft families (flap/LED/VHS/
  arcade/dossier, laser/thunder/hologram/biolume/aurora/spectrum, papercut/
  popup/chalkboard/graffiti/brush/inkwater/ransom + earlier 5 constitutions)
- themes engine: 18+ body paradigms & hero setpieces, char-widths.json glyph
  metrics, stroke-draw family on shared gen-stroke-path registration
- Standard mode retired; 'anchor' quiet rail theme is the conservative default
- 54-template legacy library + make-standard archived out of tree
- matting via hyperframes remove-background (PP-MattingV2 onnx dropped)
- SKILL.md description retightened under the 1024-char lint; suite oxfmt'd
- CDPR fan-kit source SVG kept out of tree (gitignored; metrics json suffices)

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

* fix(embedded-captions): clear CI lint — dead declarations + backtick rephrase

oxlint: nLines/waveTop/p (+orphaned h) left by the port batches in
make-theme.cjs. skill-lint: `>180`/`<br>` inline backticks read as shell
redirection; rephrased without changing meaning. Fixture regressions green
(laser/anchor/ransom recompile clean).

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

* fix(embedded-captions): read-with-catch for matte.fps (CodeQL js/file-system-race)

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

* fix(embedded-captions): e2e cold-start findings — VFR matte desync +6

Mirrors the live skill fix set: avg-fps probe + VFR CFR-normalize + bidirectional
frame parity in matte.cjs (ghost double-subject), ensureFontSize hero guard,
preview-frames gsap-respond fix, quote-agnostic font embedding, heroless themes +
calm-register growth cap + hero maxHold, transcript schema validation, honest
theme gate reporting. Verified: 19/19 fixture regression, C1/T3/T4 re-rendered.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

* docs(skills): quote frontmatter descriptions for YAML safety

Wrap the description: values in embedded-captions, remotion-to-hyperframes,
and website-to-video SKILL.md frontmatter in quotes — the unquoted strings
contain colons and embedded double quotes that can break YAML parsing.
oxfmt normalizes the two with embedded quotes to single-quoted form.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

---------

Co-authored-by: jieling-jenson <jie.ling@heygen.com>
Co-authored-by: Claude Fable 5 <noreply@anthropic.com>
2026-06-14 10:31:23 +08:00

55 KiB
Raw Blame History

THE CATALOG — one front-end, three engines behind

This is the skill's single selection surface. The user picks an IDENTITY from this table; everything else (engine, compiler, authoring file) is derived by lookup — never ask the user (or yourself) "Standard or Cinematic or Theme?" Those are backend names; a product has one UX even when it has several engines.

Routing procedure (replaces any mode decision):

  1. Probe the clip (decision gate + pre-flight probes as in SKILL.md).
  2. Shortlist 23 identities from this table by content register + the user's own words + scene constraints.
  3. Recommend ONE with a one-line why; the user picks.
  4. Author the identity's file (last column), then render with its engine's pipeline. Done — no category question ever surfaced.

Identities

Identity Reading surface Voice (one line) Recommend for Scene needs Author → engine
cream scene-embedded column flow warm poetic light, emergence + glow introspective, narrative, emotional monologues dark/mid-tone warm scenes (luma < 150) cinematic.json → make-cinematic
ink scene-embedded column flow letterpress pressed into bright surfaces the bright scenes everything else washes out on BRIGHT scenes (luma > 150) cinematic.json → make-cinematic
editorial scene-embedded column flow lowercase italic serif, pen-stroke glides fashion / beauty / lyrical / premium brand films mid/dark scenes cinematic.json → make-cinematic
keynote scene-embedded column flow confident centered slides-in-the-scene launches, founder updates, tech explainers (mood over verbatim reading) any mid/dark cinematic.json → make-cinematic
documentary scene-embedded column flow burn-in stillness, bone on charcoal interviews, journalism, testimony — gravitas IS the style any cinematic.json → make-cinematic
loud scene-embedded, body in front percussion type that competes with the face hype, social reels, sport, music, motivational any (counter-accent handles warm) cinematic.json → make-cinematic
neon scene-embedded column flow captions as neon signage in the dark cyberpunk / nightlife / tech-noir / music dark scenes cinematic.json → make-cinematic
glitch scene-embedded column flow corrupted-feed RGB-split type hacker / AI / dystopia content mid/dark cinematic.json → make-cinematic
chrome scene-embedded column flow liquid-metal hero with sheen sweeps fashion-tech, Y2K, product-flex mid/dark cinematic.json → make-cinematic
velocity scene-embedded column flow every word arrives along its motion vector sport, automotive, fitness, kinetic content any mid/dark cinematic.json → make-cinematic
anchor verbatim lower-third rail + 1 settled climax the quiet professional — fade-up rail that simply reads; the apex settles behind the subject with a drawn accent underline. Discipline IS the style. explainers, tutorials, corporate, news-adjacent — where every word must be read; the conservative default any (opaque rail; settle carries a soft veil) theme.json → make-theme
ordnance stamped verbatim rail + detonation apex military marking system; everything has mass; the room reacts punchy announcements, drops, "炸一点 / 像 AE 做的" asks any (charge-dim makes its own contrast) theme.json → make-theme
terminal typed console panel + glyph-decode apex the transcript is a signal log; apex decodes from noise tech / data / AI / investigative, "终端/控制台/解码" asks cool/dusk preferred theme.json → make-theme
neonsign warm tube rail + hand-written neon apex a sign district at dusk; the apex is WRITTEN stroke by stroke lifestyle / music / mood, "霓虹/手写" asks dusk/dark backdrop; needs python3 (stroke-path gen) theme.json → make-theme
stardust accumulating poem, condensing letters words condense from dust and return to it poetic / memory / science-wonder monologues one clear column side (probe fails → fall back to cream) theme.json → make-theme
stomp full-frame takeover cards hard-cut typographic takeover with a void before the drop hooks, manifestos, ad openers — the words ARE the show any (plate runs dimmed) theme.json → make-theme
lastpage manuscript rail + a blurred future-field the future hangs unreadable in the room; one rack-focus reveals every faint line was the apex word visionary thesis, prophecy-toned talks, trailers, literary content mid/dark (the field needs air) theme.json → make-theme
nightcity corrupted-UI rail + cover-letterform apex body words boot through a corrupted-cyan frame into acid-yellow UI; the apex slams in BEHIND the subject set in the CP2077 cover replica typeface (cyan offset, baseline streak, circuit trace), tears in as displaced slices, lives as a glitching print gaming / tech-edge / glitch culture / hype drops, "赛博朋克/游戏封面" asks any (the lockup carries its own contrast); CP2077-adjacent content only (fan-kit terms) theme.json → make-theme
scoreboard split-flap chyron rail + departures-board apex stadium split-flap mechanics — body words flip-cycle wrong words on tiles before clacking onto the real one; ROW counter + odometer chip furniture; the apex is a giant departures board behind the subject that locks the word letter-by-letter under floodlights and a crowd-flash, status ticker types on sports, transit/travel, scores, countdowns, reveals, "翻牌/记分牌/机场大屏" asks any (board + chyron bring their own dark metal contrast) theme.json → make-theme
transit LED departure-strip rail + illumination-wipe apex amber dot-matrix station PA — body words light column-by-column (L→R LED wipe + arrival flicker) on a docked departure strip whose lines page by scroll-snap; platform tag, ticking clock, ⚠ DELAYED → ● ON TIME gag; the apex panel expands behind the subject inside a chasing LED-marquee border, the word illuminates column-by-column under a PA flash + ding pulses, then pages up and collapses travel/transit, announcements, schedules, countdowns, retro-terminal, "LED屏/车站到站/点阵屏" asks any (LED strip + panel bring their own dark contrast) theme.json → make-theme
vhs camcorder OSD rail + REC-FREEZE tracking-slam apex tape playback — permanent lens vignette + OSD furniture (▶ PLAY, ● REC, blinking green timestamp); body words carry a permanent red/cyan misregistration fringe, arrive by tracking glitch and REWIND off with flat-white echoes + scrub lines (the deck flips PLAY→REW in dead air); the apex is REC FREEZE: 1-frame white tear + head-switching noise bar, the word slams in behind the subject as three sheared tracking bands that SNAP into register, misregistration breathes through the hold under tape weave + auto-iris, then rewinds off retro / nostalgia / found-footage / home-video, '90s tech, music & lifestyle edits, "录像带/VHS/复古/DV" asks any (backlight-comp scrim makes its own contrast on bright skies) theme.json → make-theme
arcade arcade HUD rail + pixel-block boss-intro apex 8-bit cabinet attract mode — the frame plays through CRT scanlines + a corner vignette; body words coin-pop onto a bordered HUD bar (blinking 1UP, CREDIT 01) with a 1-frame white blip each, emphasis words run a cyan→magenta palette swap, lines dissolve checkerboard; the apex is a BOSS INTRO behind the subject: the room dims, WARNING blinks twice, ~64 seeded pixel blocks fall quantized into a matrix grid over the word rect, the word flashes ON in Press Start 2P with chromatic ghost layers + a 3-frame palette strobe while letterbox bars slam in and a HI-SCORE odometer rolls top-right; everything dissolves checkerboard and the blocks drop away gaming, retro-tech, playful hype, launches, achievement/level-up beats, "像素/街机/8-bit/打游戏" asks any (boss dim + dark HUD bar bring their own contrast) theme.json → make-theme
dossier manila case-file strip + rubber-stamp apex cold-war case file — carbon Special Elite HAMMERS onto a manila transcript strip char by char (seeded baselines + ribbon ink, every strike kicks the paper; typed case header, CONFIDENTIAL watermark, red ink circles around the evidence; pages are X-row over-struck and carriage-fed away); the apex is a RUBBER STAMP behind the subject: the footage goes sepia-archival, the stamp crushes in at 8° with a double-strike ghost + thin-ink patches, then holds DEAD STILL — sold only by a breathing paper indent — until the page is filed away with a mechanical yank investigations, exposés, true-crime, document-driven storytelling, declassified/archival framing, "档案/卷宗/解密/盖章" asks any (the manila strip is self-lit; the stamp brings its own sepia charge) theme.json → make-theme
laser beam-ignited rail + 6-beam laser-cage apex concert laser rig in a blackout — every body word is IGNITED where a green and a magenta beam converge on it two frames early (flare pop, snap-on, single strobe on emphasis; a sweeper beam carries each line away); the apex is the drop: a 6-beam array sweeps to one hot point, the word ignites behind the subject in an additive glow stack with two controlled strobe pops, a criss-cross beam cage draws around it, the array fans out wide through the hold, then the beams sweep everything off L→R music, nightlife, festivals, drops, hype reveals, keynote 'one more thing' beats, "激光/灯光秀/蹦迪字幕" asks any (the concert blackout makes its own contrast, bright skies included) theme.json → make-theme
thunder rain-washed rail + branched-bolt strike apex the storm is the editor — slate storm mood on the room, ambient slanted rain falling in front of the subject, and the sky "kisses" (1-frame brighten) on every body word; rail words arrive with a lit-bright lightning pop (emphasis = sheet-lightning double flash mirrored by a double sky-kiss); the apex is THE STRIKE: a hand-authored branched bolt flashes two frames before the word, the word slams in lit HARD with its shadow flipped toward the bolt, ozone under-glow breathes through the hold while the rain doubles, then everything washes downward with the rain dramatic reveals, storm/weather content, epic hype beats, trailer-toned monologues, "雷电/闪电/暴风雨字幕" asks any (charge-dim + radial word scrim make their own contrast, bright skies included) theme.json → make-theme
hologram volumetric projection plate + hologram-boot apex every caption is LIGHT cast from an emitter — body words boot onto a scanlined projection plate with a rising sweep, a 1-frame interference glitch and a permanent cyan/magenta fringe (emphasis words drop out for 2 frames and restore brighter; lines drift in rotationY and exit by projector cut); the apex is the full volumetric boot: a desk emitter dot pops, the cone snaps wide, a seed beam fires, and the word materializes behind the subject as a 7-layer hologram with climbing interference bands — a hand passes through it before the projector cuts it back to the dot tech / sci-fi / AI / space / imaging explainers, futuristic product reveals, "全息/投影/科幻字幕" asks any (the apex scrim + dim make their own contrast; wants a lower frame edge for the emitter dots) theme.json → make-theme
biolume jellyfish glow-on rail + bioluminescent-bloom apex deep-sea bioluminescence — the clip sinks into an abyss grade with a breathing vignette and every caption is a living light: body words GLOW ON like jellyfish (radial bloom from center while a cyan glow breathes in; emphasis words pulse a slow propulsion double-beat) across two coexisting floating rows and SINK away dimming; seeded plankton motes drift the open water and the ones ringing the apex are pulled into the bloom; the apex is the BLOOM behind the subject: a creature pre-glow double-beats and crushes to a point, the word condenses in, six luminous tendrils draw outward like filaments while caustics shimmer, cyan light spills onto the scene and spores burst past — then all the light sinks back into the deep ocean / nature / science-wonder docs, dreamy or meditative monologues, slow night-time reveals, "深海/发光水母/水下字幕" asks any (the abyss grade + pocket scrim make their own contrast on bright scenes) theme.json → make-theme
aurora iridescent sheen rail + silk-ribbon draw-on apex polar-night elegance — two huge aurora blobs drift the sky all clip; body words are silk serif ink whose rose→teal→violet sheen drifts continuously through the letters (entrance = 2-frame flow-on + y8 settle; emphasis = one bright highlight sweep marching through the word), lines share two coexisting rows like a woven couplet and dissolve UPWARD into drifting light streaks; the apex is a SILK RIBBON: the hero word is hand-WRITTEN into the sky behind the subject stroke by stroke (script single-line font, blurred halo + bright core, a light mote riding the pen) and the aurora gradient keeps FLOWING through the drawn ribbon during the hold while a soft spill tints the sky and light motes rise past the subject — gentle, no shake dreamy / romantic / wonder content, night-sky and travel pieces, poetry, weddings, elegant brand films, "极光/丝带/优雅手写字幕" asks any (apex scrim + dark understroke carry contrast on bright skies; loves dusk/night) theme.json → make-theme
spectrum oscilloscope scope-band rail + phosphor trace-write apex an oscilloscope is listening — a dark scope band hosts an ALWAYS-ALIVE waveform (seeded pseudo-RMS envelope: speech grows it, silence flattens it, every word arrival kicks a local burst) under VT323 phosphor lines that wipe in L→R over reserved slots with a 2-frame just-traced flash (emphasis words get a ticking frequency-readout chip, the trig LED blinks on every word); the apex is the trace itself LEAPING off the scope (surge column) to hand-WRITE the hero word in a thin phosphor script stroke, fill flicker-on, hum + breathe, then COLLAPSE to a flatline before the next spoken word — and on the FINAL word the trace explodes to max-amp chaos, flatlines to a single dot and stamps SIGNAL LOST science / signal / audio / measurement / astronomy explainers, data-nerd content, retro-lab or synth aesthetics, "示波器/波形/信号字幕" asks any (the scope band brings its own dark glass; the apex scrim handles bright skies); needs python3 (stroke-path gen) theme.json → make-theme
papercut paper-chip craft rail + stop-motion papermat apex stop-motion paper craft — the frame becomes an animator's table: body words are cream paper chips PLACED one by one (steps(2) drop, seeded crooked, ~5fps wobble) on a kraft strip with torn top edge, washi-taped corners and a handwritten cut&paste tag; emphasis words get a washi-tape slap, the final word is an accent-red sign-off chip, and a red paper star stands in for the hero word; the apex is a torn kraft mat slapped down behind the subject where torn-edge letter chips DROP with stepped gravity + anticipation shadows, land crooked, settle in 12fps poses, wobble at 8fps while a warm vignette charges — then the FINAL word bumps the table and everything scatters off down-right craft/DIY, kids & family, indie product demos, behind-the-scenes, maker channels, "手工/剪纸/定格动画字幕" asks any (kraft strip + cream chips bring their own warm contrast) theme.json → make-theme
popup pop-up book page rail + centerfold starburst apex a pop-up book opens inside the frame — body words POP UP on a baseline hinge (rotationX back.out, paper overshoot, a growing back-shadow) along a cream page strip that unfolds from the bottom edge (center foldline, two alternating reading rows); emphasis words land bigger with a contact squash the page FEELS (1px strip dip), then flick-wobble with a blue paper strut leaning in behind; line exits fold flat in an L→R wave (the page closing), the whole page TILTS on the final word and the strip folds shut on the last frame; the apex is THE CENTERFOLD: blue + cream 12-point paper starbursts unfold behind the subject, the hero word pops up in three hinged clip-path segments, kraft struts lean in, backdrop + word wobble-settle TOGETHER and breathe through the hold — then the whole spread folds flat before the next sentence; a blue paper star holds the hero slot on the rail and paper confetti flicks past the subject at the pop kids and family content, storytelling, education, greeting-card / wholesome brand spots, picture-book and craft channels, "立体书/弹出书/纸艺字幕" asks any (the cream page strip is self-lit; the apex dusk scrim guards bright skies) theme.json → make-theme
chalkboard chalk-band rail + hand-written chalkwrite apex night lecture — a worn deep-green chalk band where every word is SCRIBBLED in fast chalk (2-frame pop + steps(4) scribble reveal + dust puff at the base, seeded crooked); emphasis gets double underline dashes, the final word lands in yellow chalk, and between sentences an ERASER SWIPE smears the old words off (blurred light band, ghost smudges stay — chalk never fully erases); the apex: the band GROWS into a full blackboard behind the subject and the hero word is HAND-WRITTEN 130px tall stroke by stroke (jittered chalk nib, dust falling from the pen, blurred halo under a tight core), finished with a hand-slap dust CLAP in front of the subject + a 1-frame board shake and two violent accelerating underlines, then the halo breathes through the hold education / explainers, science and lecture content, teacher channels, documentary asides, "黑板/粉笔/板书/老师讲课字幕" asks any (the deep-green board brings its own contrast); needs python3 (stroke-path gen) theme.json → make-theme
graffiti marker rail + spray-tagged handstyle apex street bombing run — marker words FLICK onto the rail crooked (1-frame squeak skew, seeded rotation, back.out settle); emphasis takes alternating yellow/magenta paint + a spray-underline swoosh with overspray dots; spent lines are ROLLER-BUFFED away (gray roller sweep, wet smear lingers) and the final word lands BIG in yellow with the punch; the apex: the dusk wall dims, the can RATTLES, and the hero word is spray-TAGGED stroke by stroke behind the subject in fat yellow paint (dark under-stroke armor + blurred overspray halo + glowing nib + seeded overspray dots), paint DRIPS grow from the low points through the hold, a magenta paint splat pops at the pen-finish and the plate punches on it street culture, hip-hop / rap, sneaker & fashion drops, urban sports, skate/BMX, rebellious brand spots, "涂鸦/喷漆/街头字幕" asks any (the dark under-stroke armors the tag against bright skies); needs python3 (stroke-path gen) theme.json → make-theme
brush washi paper-band rail + sumi-e brush-gesture apex sumi-e study — a deckle-edged washi paper band swipes in behind a traveling brush-head shadow; body words are brush-WIPED on in a Mincho serif (clip reveal behind a per-word brush head, seeded ±1° tilt), emphasis gets a double curved ink underline, and spent lines are wiped OFF right-to-left like a wet cloth (erase brush head riding the wipe); the apex: a warm paper wash charges behind the subject and the hero word is written in ONE violent calligraphic gesture — pressure-tapered wet brush (four stacked stroke widths, blurred bleed underneath), start-SPLAT at the first pen contact with ink flecks kicking past the subject, spatter droplets ejected at the sharpest curves — then the ink BLEEDS into the paper through the hold (the work breathes) and a vermilion SEAL chop (観) stamps beside the ink with the plate punch calm-authority explainers, craft / design / philosophy / Japan-adjacent content, tea / meditation / writing channels, elegant brand films, "水墨/毛笔/书法字幕" asks mid/bright scene band (dark ink + its paper wash carry the contrast); needs python3 (stroke-path gen) theme.json → make-theme
inkwater three-row ink rail + ink-drop bloom apex sumi ink dropped in still water — paper-white radial washes back the text zones, Mincho-serif ink words BLEED into three fixed reading rows (blur+scale bleed over a paper-glow halo, dark blurred halo overshoots then tightens, an ink-settling hairline under each line, emphasis sheds a curling ink wisp) and spent lines re-dissolve upward like dispersing ink; the apex: a dark drop falls behind the subject, hits an invisible water surface at the word point and BLOOMS (three stacked blurred ink ellipses at counter-rotating rates + two thin ripple rings) while the world stills under a gentle vignette — the hero words resolve out of the blob on their own spoken times, breathe at the glyph edges through the hold, then re-dissolve before the next thought needs the reader contemplative / literary / scientific-wonder narration, astronomy / nature / memory / essay films, poetry and slow-craft channels — one liquid meditative event instead of an impact, "水墨滴落/墨水字幕" asks mid/bright scene band (dark ink + its paper washes carry the contrast) theme.json → make-theme
ransom cutout-chip collage rail + ransom-note slam apex kidnapper's collage — every word is a chip clipped from a different page (cream Anton / black Inter / newsprint typewriter / accent-yellow marker, torn edges, hard offset shadows) PASTED crooked on the rail with misaligned baselines (the apex slam at low amplitude); emphasis words get a 1-frame RE-PASTE lift (shadow grows, rotation flips sign for good) and spent chips are RIPPED off one by one under 1-frame paper-tear flashes; the apex: a collage scrim darkens the sky and the hero word slams in letter by letter as a ransom note BEHIND the subject — each letter a different cutout chip, flight shadow popping tight on contact — the whole word JOLTS as the last letter lands (the plate punches with it), looms crooked through the hold, then is torn off chip by chip while the closing line speaks true crime, mystery / thriller, heist energy, punk / DIY zine aesthetics, edgy drops, unhinged-comedy hooks, "勒索信/剪贴/拼贴字幕" asks any (solid chip backgrounds carry their own contrast; the apex scrim guards bright skies) theme.json → make-theme

Shortlisting heuristics (identity-level, not category-level)

  • "每个词都要被读到 / accessibility / 教程·新闻" → rail or panel surfaces: anchor, ordnance, terminal (all verbatim by construction).
  • "嵌进场景 / 电影感 / 高级" → the column-flow ten; pick by register + scene luma (bright → ink; warm dark → cream; fashion → editorial; …). Bright + must-read → anchor.
  • "炸 / 特效 / 酷炫 / 像 AE"ordnance, stomp, terminal, loud — pick by what should explode: a word (ordnance), the frame (stomp), a decode (terminal), every beat (loud).
  • Named worlds → direct hits: 控制台/终端 → terminal; 霓虹/招牌 → neonsign; 诗/星尘/记忆 → stardust; 军事/重击 → ordnance; 预言/未来/悬念揭示 → lastpage; 赛博朋克/游戏封面/夜之城 → nightcity.
  • Scene constraints veto last: luma > 180 kills cream/screen column looks (→ ink or a rail/panel surface); dusk/dark unlocks neon/neonsign.
  • Unsureanchor (the conservative default: words read, scene safe, zero spectacle). EXCEPTION: if the ask literally says "嵌进场景 / embed into the scene / 电影感", that signal OUTRANKS the conservative default — go column-flow (pick by register + luma), not a rail.

Adjacencies worth knowing (these pairs are why the old mode question was hard — recommend by the difference, don't agonize over category): loud vs ordnance (percussive type vs a themed world with plate reaction) · neon vs neonsign (neon-styled captions vs a sign being written) · cream vs stardust (warm embedded flow vs an accumulating poem with semantic dispersal) · terminal vs glitch (a console WORLD with typed log + decode vs corrupted-feed styling on normal captions) · stomp vs loud (every beat takes over the dead-center frame vs energetic type that still shares the frame with the subject).

Rules

  • Identities are engine-locked — no cross combos. Opening a new combo (e.g. another DNA under rail) is a validation event: render → user review → flip the DNA's deliveries.rail flag (see dna/README.md).
  • New identities join via their family's registry rules (dna/README.md for classic, themes/README.md for themed — "any input, zero hand-fixing" bar).
  • The engines' internals (Standard/Cinematic/Theme compilers, gates, layers) are documented in their own files; this catalog is the only routing surface.