Self-hosted, Self-Improving, Federated Everything App — dev management, agent orchestration, persistent mind, digital twin, second brain, and creative pipeline from a single dashboard over Tailscale to port the user into the OS
JavaScript ai-agentsdashboarddeveloper-toolsdigital-twinquantified-selfsecond-brainself-hostedtailscale
Overlay text check: sample the first frame of every word
Claude9eaf6e79669e
diff --git a/server/services/musicVideo/overlayText.js b/server/services/musicVideo/overlayText.jsindex 2dfe517044..0e366082af 100644--- a/server/services/musicVideo/overlayText.js+++ b/server/services/musicVideo/overlayText.js@@ -25,6 +25,8 @@ import { createHash, randomUUID } from 'node:crypto';const PHONE_WIDTH_PX = 390;const OVERLAY_TEXT_LIMITS = Object.freeze({maxSamples: 180,+ // Entrance frames: the first frame of each line and timed word, where a slam-in is at its largest.+ maxEntrances: 240,maxFindings: 40,maxTimesPerFinding: 8,// Smallest em (font size) still readable at phone width, in phone pixels.@@ -86,11 +88,18 @@ const contrastRatio = (a, b) => (Math.max(a, b) + 0.05) / (Math.min(a, b) + 0.05* Song times (seconds, on the frame grid) to sample: each lyric line and text* cue once settled (0.35s after it starts), mid-way through a long one and near* its end; each shot just after it starts and at its middle. Times closer than- * 0.15s collapse; more than `maxSamples` thin evenly.+ * 0.15s collapse; more than `maxSamples` thin evenly. On top of those, the+ * first frame each line, cue and timed word is on screen: a word that slams or+ * stamps in is drawn largest there, so a transient swell past its neighbours or+ * the frame edge lasts only a frame or two. Those are kept frame-exact (up to+ * `maxEntrances`, thinned evenly).*/-export function planTextSampleTimes({ lyrics = [], textCues = [], scenes = [], durationSec, fps = 24, maxSamples = OVERLAY_TEXT_LIMITS.maxSamples }) {+export function planTextSampleTimes({ lyrics = [], textCues = [], scenes = [], durationSec, fps = 24,+ maxSamples = OVERLAY_TEXT_LIMITS.maxSamples, maxEntrances = OVERLAY_TEXT_LIMITS.maxEntrances }) {if (!(durationSec > 0)) return [];const times = [];+ const entrances = [];+ const enter = (t) => { if (finite(t)) entrances.push(t); };const span = (start, end) => {if (!finite(start)) return;const stop = finite(end) && end > start ? end : start + 1;@@ -103,8 +112,10 @@ export function planTextSampleTimes({ lyrics = [], textCues = [], scenes = [], dconst words = Array.isArray(line?.words) ? line.words.filter((w) => finite(w?.startSec)) : [];const end = finite(line?.endSec) ? line.endSec : words.length ? Math.max(...words.map((w) => (finite(w.endSec) ? w.endSec : w.startSec))) : null;span(line?.startSec ?? words[0]?.startSec, end);+ enter(line?.startSec);+ for (const w of words) enter(w.startSec);}- for (const cue of Array.isArray(textCues) ? textCues : []) span(cue?.startSec, cue?.endSec);+ for (const cue of Array.isArray(textCues) ? textCues : []) { span(cue?.startSec, cue?.endSec); enter(cue?.startSec); }for (const scene of Array.isArray(scenes) ? scenes : []) {if (!finite(scene?.startSec) || !finite(scene?.endSec) || scene.endSec <= scene.startSec) continue;times.push(scene.startSec + Math.min(0.25, (scene.endSec - scene.startSec) / 2));@@ -116,10 +127,16 @@ export function planTextSampleTimes({ lyrics = [], textCues = [], scenes = [], d.sort((a, b) => a - b);const kept = [];for (const t of snapped) if (!kept.length || t - kept[kept.length - 1] >= 0.15) kept.push(round3(t));- if (kept.length <= maxSamples) return kept;- return Array.from({ length: maxSamples }, (_, i) => kept[Math.floor(((i + 0.5) * kept.length) / maxSamples)]);+ // The first frame drawn at or after the entrance (a frame before it shows nothing yet).+ const firstFrames = [...new Set(entrances+ .map((t) => round3(Math.min(last, Math.max(0, Math.ceil(t * fps - 1e-6) / fps)))))].sort((a, b) => a - b);+ const merged = new Set([...thinEvenly(kept, maxSamples), ...thinEvenly(firstFrames, maxEntrances)]);+ return [...merged].sort((a, b) => a - b);}+const thinEvenly = (list, max) => (list.length <= max ? list+ : Array.from({ length: max }, (_, i) => list[Math.floor(((i + 0.5) * list.length) / max)]));+const widthOf = (b) => b.x1 - b.x0;const heightOf = (b) => b.y1 - b.y0;const emOf = (item) => (finite(item.emPx) && item.emPx > 0 ? item.emPx : heightOf(item));@@ -282,10 +299,10 @@ const quote = (text) => `“${text}”`;/** One plain sentence for a finding (what is wrong, and the fix lyricType uses). */function describeTextFinding({ kind, texts, detail }) {switch (kind) {- case 'overlap': return `${quote(texts[0])} collides with ${quote(texts[1])}. Move one to a free corner or time them apart.`;+ case 'overlap': return `${quote(texts[0])} collides with ${quote(texts[1])}. Move the smaller one to the corner opposite the sung words, or time them apart.`;case 'off-frame': return `${quote(texts[0])} runs off the edge of the frame. Keep it inside the safe margin or let it wrap.`;case 'small': return `${quote(texts[0])} is ${detail?.phoneEmPx ?? 'too few'}px tall at phone width. Set it larger (at least ${OVERLAY_TEXT_LIMITS.minPhoneEmPx}px on a phone).`;- case 'contrast': return `${quote(texts[0])} blends into the picture (${detail?.ratio ?? '?'}:1). Give it an ink outline wide enough to read on a phone${detail?.outlined ? ' (its outline is too thin there)' : ''}.`;+ case 'contrast': return `${quote(texts[0])} blends into the picture (${detail?.ratio ?? '?'}:1). Give it an ink outline wide enough to read on a phone${detail?.outlined ? ' (its outline is too thin there)' : ''}, or a plate behind it.`;default: return quote(texts.join(' / '));}}diff --git a/server/services/musicVideo/overlayText.test.js b/server/services/musicVideo/overlayText.test.jsindex 91691859f2..f75a6026d2 100644--- a/server/services/musicVideo/overlayText.test.js+++ b/server/services/musicVideo/overlayText.test.js@@ -80,13 +80,14 @@ describe('summarizeTextFindings', () => {});describe('planTextSampleTimes', () => {- it('samples each line settled, long lines again, and each shot, on the frame grid without near-duplicates', () => {+ it('samples each line settled, long lines again, each shot, and the first frame of each word, on the frame grid', () => {const times = planTextSampleTimes({- lyrics: [{ startSec: 1, endSec: 4, words: [] }],+ lyrics: [{ startSec: 1, endSec: 4, words: [{ startSec: 1 }, { startSec: 2.02 }] }],scenes: [{ startSec: 0, endSec: 8 }],durationSec: 8, fps: 24,});- expect(times).toEqual([0.25, 1.333, 2.5, 3.792, 4]);+ // 1 and 2.042 are entrance frames (a slam-in is largest there), kept however close to a settled sample.+ expect(times).toEqual([0.25, 1, 1.333, 2.042, 2.5, 3.792, 4]);expect(times.every((t) => Math.abs(t * 24 - Math.round(t * 24)) < 0.05)).toBe(true);const many = planTextSampleTimes({ scenes: Array.from({ length: 300 }, (_, i) => ({ startSec: i, endSec: i + 1 })), durationSec: 300, maxSamples: 50 });expect(many).toHaveLength(50);