@wji.ai/cli 0.8.2 → 0.9.0

package/skill/short-drama/reference/workspace-layout.md

@@ -73,7 +73,7 @@ the headshot + sheet as image references and the voice as audio.
 ```
 
 - headshot: `canvas add image -n "C01 headshot" -m <image model> -r 1:1 -p "<image.prompt>"`
-- sheet:    `canvas add image -n "C01 sheet" --from <headshotNode> -m <image model> -r 16:9 -p "@image1 is the face to match. <image.sheet>"`
+- sheet:    `canvas add image -n "C01 sheet" --from <headshotNode> -m <image model> -r 16:9 -p "@image1 is the face to match exactly. <image.sheet>"`
 - voice:    `canvas add audio -n "C01 voice" -m <audio/tts model> -p "<voice.prompt>"` (optional; skip when no voice model is enabled)
 - title:    `canvas add title -t "C01 林小夏" --at <x>,<y>` — one per row, rows 700 px apart
 

package/skill/short-drama/scripts/selftest.mjs

@@ -14,7 +14,7 @@ import {
   paramsOf, parseEpRange, promptSkeleton, renderCharacter, renderOutline, renderProp, renderScene, renderScript,
   renderSegment, runGates, seedScript, seedStoryboard, seedanceAnchors, slug, timestamp, writeJsonFile,
   wanLegendLine, wanSheetLegendLine, wanShotHeader, segmentReferences, imageReferences, audioReferences, voiceSamplePrompt, voiceSampleSeconds, nonWhitespace,
-  speechDirection, cameraPhrase, lightPhrase, foregroundPhrase, focalOf, LIGHT_DIRECTIONS, identityOf, isEnglish,
+  speechDirection, cameraPhrase, lightPhrase, idlePhrase, idleSentence, IDLE_MAX_CHARS, eyelinePhrase, eyelineValid, onlyInFramePhrase, overShoulderPhrase, sidesLine, foregroundPhrase, foregroundBand, foregroundForLook, focalOf, LIGHT_DIRECTIONS, identityOf, isEnglish,
   renderGuideline, renderDialogue, scriptSelfCheck, beatOffsets, turnPoints, parseAge, talkAndAction, normalizeRegister, REGISTERS, REGISTER_ACCENT,
   cutStart, segmentCharacters, keyScenes, shotTexts, motionSentence, cameraClause, PURPOSE_KINDS, ANGLES, HEIGHTS, NO_SUBTITLES, LIPS_CLOSED,
   audioPrompt, musicOf, cueSeconds, AUDIO_MODEL, AUDIO_MAX_CHARS, AUDIO_MAX_SECONDS, scoreSheet, smpte,
@@ -850,18 +850,127 @@ const breaks = (gateId, project, episode, needle) => {
   breaks('photography', PROJECT, noNote, 'light.note is required');
   const wrongDir = clone(E01); wrongDir.storyboard.segments[0].cuts[0].light.direction = 'top';
   breaks('photography', PROJECT, wrongDir, 'keyed from directly above');
-  // Foreground: named in the frame, said to be out of focus, and present on at least half the cuts.
+  // Foreground: named in the frame, said to be out of focus, and within params.foregroundLevel across the episode.
   ok(foregroundPhrase(cut).includes('out of focus'), 'the foreground phrase says it is defocused');
   eq(foregroundPhrase({ foreground: '' }), '', 'a clean frame adds nothing');
   ok(foregroundPhrase({ foreground: '\u6536\u94f6\u53f0' }, 'zh').includes('\u524d\u666f'), 'the Chinese foreground phrase says 前景');
-  const noFg = clone(E01); noFg.storyboard.segments[0].cuts.forEach((c) => { c.foreground = ''; });
-  breaks('photography', PROJECT, noFg, 'no depth');
+  const noFg = clone(E01); noFg.storyboard.segments.forEach((seg) => seg.cuts.forEach((c) => { c.foreground = ''; }));
+  // The fixture sits at `high`: no foreground at all is too few there, and fine at `low`.
+  breaks('photography', PROJECT, noFg, 'params.foregroundLevel high allows');
+  const asLevel = (level) => ({ ...PROJECT, params: { ...PROJECT.params, foregroundLevel: level } });
+  ok(!failures('photography', asLevel('low'), noFg).some((f) => f.includes('foregroundLevel')), 'low accepts an episode of clean frames');
+  breaks('photography', asLevel('low'), E01, 'params.foregroundLevel low allows');
+  breaks('photography', asLevel('medium'), E01, 'params.foregroundLevel medium allows');
+  breaks('params', asLevel('sometimes'), E01, 'params.foregroundLevel must be one of');
+  const levelFor = (params) => paramsOf({ params }).foregroundLevel;
+  eq(levelFor({}), 'medium', 'the default level is medium');
+  eq(levelFor({ composition: 'standard', cinematography: 'low-key', genre: '复仇' }), 'medium', 'a standard look stays medium');
+  eq(levelFor({ composition: 'art-first' }), 'high', '艺术优先 defaults to high');
+  eq(levelFor({ cinematography: 'sweet-romance' }), 'high', 'the 唯美 romance preset defaults to high');
+  eq(levelFor({ cinematography: 'costume-idol' }), 'high', 'the 唯美 costume preset defaults to high');
+  eq(levelFor({ genre: '唯美古风' }), 'high', 'a 唯美 genre defaults to high');
+  eq(levelFor({ genre: '美感優先' }), 'high', 'a 美感優先 genre defaults to high');
+  eq(levelFor({ composition: 'art-first', foregroundLevel: 'low' }), 'low', 'a named level beats the look');
+  eq(foregroundForLook({ cinematography: 'costume-dark' }), 'medium', 'the dark costume preset is not 唯美');
+  eq(foregroundBand('low', 40).max, 10, 'low allows a quarter of the cuts');
+  eq(foregroundBand('medium', 40).min, 10, 'medium starts at a quarter');
+  eq(foregroundBand('medium', 40).max, 20, 'medium stops at half');
+  eq(foregroundBand('high', 40).min, 20, 'high starts at half');
+  ok(foregroundBand('nope', 40) === null, 'an unknown level has no band');
+  for (const [k, pr] of Object.entries(CINEMATOGRAPHY_PRESETS)) if (pr) ok(!/foreground/i.test(pr.videoPhrase) && !/前景/.test(pr.videoPhraseZh), `${k} leaves the foreground to the level`);
   const missingFg = clone(E01); delete missingFg.storyboard.segments[0].cuts[0].foreground;
   breaks('photography', PROJECT, missingFg, 'foreground is required');
   const unsaidFg = clone(E01); unsaidFg.storyboard.segments[0].cuts[0].frame = unsaidFg.storyboard.segments[0].cuts[0].frame.replace('the dripping edge of the striped awning', 'a wall');
   breaks('photography', PROJECT, unsaidFg, 'must name the foreground');
 }
 
+{
+  // Everyone on screen is alive: a model freezes whoever the prompt gives no motion.
+  eq(idlePhrase('shifts her weight', '@image3'), '@image3 shifts her weight.', 'an idle reads as a sentence');
+  eq(idlePhrase('still', '@image3'), '@image3 holds still, only breathing and blinking.', 'still keeps the breath');
+  ok(idlePhrase('轻轻换了一下重心', '@image3', 'zh').endsWith('轻轻换了一下重心。'), 'the Chinese idle');
+  eq(idleSentence({ characters: ['A', 'B'], idle: { B: 'taps a finger' } }, (id) => id), 'B taps a finger. ', 'only the people with an idle are written');
+  const p = clone(PROJECT); p.params.references = 'sheets';
+  const only = (f) => f.filter((m) => m.startsWith('E01-02'));
+  const build = (idle) => {
+    const ep = clone(E01); const seg = ep.storyboard.segments.find((s) => s.id === 'E01-02');
+    const two = seg.cuts[2];   // C01 + C02
+    if (idle) two.idle = idle;
+    seg.prompt = promptSkeleton(p, ep, seg.id).video.prompt;
+    return { ep, seg, two };
+  };
+  const bare = build();
+  ok(only(failures('idle-motion', p, bare.ep)).some((m) => m.includes('have an idle')), 'a second visible person with no idle fails');
+  const good = build({ C02: 'shifts her weight and picks at her sleeve' });
+  eq(only(failures('idle-motion', p, good.ep)).filter((m) => !m.includes('cameraNote.hold')).length, 0, 'one idle for the second person passes');
+  ok(good.seg.prompt.includes('shifts her weight and picks at her sleeve.'), 'the skeleton writes the idle');
+  const stillOne = build({ C02: 'still' });
+  ok(stillOne.seg.prompt.includes('holds still, only breathing and blinking.'), 'still is written with the breath');
+  eq(only(failures('idle-motion', p, stillOne.ep)).filter((m) => !m.includes('cameraNote.hold')).length, 0, 'keeping still is allowed when it is the performance');
+  ok(only(failures('idle-motion', p, build({ C02: 'walks to the door' }).ep)).some((m) => m.includes('too big')), 'a change of place is not an idle');
+  ok(only(failures('idle-motion', p, build({ C02: 'says something quietly' }).ep)).some((m) => m.includes('too big')), 'a line is not an idle');
+  ok(only(failures('idle-motion', p, build({ C02: 'x'.repeat(IDLE_MAX_CHARS + 1) }).ep)).some((m) => m.includes('one small motion')), 'an idle stays small');
+  const unwritten = build({ C02: 'taps a finger' }); unwritten.seg.prompt = unwritten.seg.prompt.replace(/taps a finger\.\s*/, '');
+  ok(only(failures('idle-motion', p, unwritten.ep)).some((m) => m.includes('must carry the idle')), 'an idle left out of the prompt fails');
+  const onlyHold = build({ C02: 'taps a finger' }); onlyHold.two.cameraNote = { ...onlyHold.two.cameraNote, hold: 'only the sign flickers' };
+  ok(only(failures('idle-motion', p, onlyHold.ep)).some((m) => m.includes('drop "only"')), '"only the sign" contradicts the people moving');
+  const offScreen = build({ C03: 'taps a finger' });
+  ok(only(failures('idle-motion', p, offScreen.ep)).some((m) => m.includes('not on screen')), 'an idle for someone off screen fails');
+  eq(failures('idle-motion', PROJECT, E01).length, 0, 'keyframes mode is exempt');
+}
+
+{
+  // Two people on screen, sheets mode: who is in frame, where they look, which side they stand on.
+  eq(eyelinePhrase('screen-right down', '@image3'), '@image3 looks toward screen-right and down.', 'an eyeline has a direction and a height');
+  eq(eyelinePhrase('task', '@image3'), '@image3 keeps their eyes on what their hands are doing.', 'task keeps the eyes on the work');
+  ok(eyelinePhrase('away screen-left', '@image3').includes('looks away'), 'away says so');
+  ok(eyelinePhrase('screen-left up', '@image3', 'zh').includes('左侧上方'), 'the Chinese eyeline');
+  ok(eyelineValid('screen-left level') && eyelineValid('task') && eyelineValid('away screen-right down'), 'valid eyelines');
+  ok(!eyelineValid('at him') && !eyelineValid('screen-right sideways'), 'an eyeline without a direction is refused');
+  const p = clone(PROJECT); p.params.references = 'sheets';
+  const framed = () => {
+    const ep = clone(E01);
+    const seg = ep.storyboard.segments.find((s) => s.id === 'E01-02');   // C01 | C02 | C01+C02 | C02 | C02, C02 screen left
+    const look = { C01: 'screen-left level', C02: 'screen-right level' };
+    seg.cuts.forEach((c) => { c.eyeline = Object.fromEntries((c.characters || []).map((id) => [id, look[id]])); });
+    seg.prompt = promptSkeleton(p, ep, seg.id).video.prompt;
+    return { ep, seg };
+  };
+  const only = (f) => f.filter((m) => m.startsWith('E01-02'));
+  const { ep: okEp, seg: okSeg } = framed();
+  eq(only(failures('two-person-framing', p, okEp)).length, 0, 'the skeleton writes everything the gate asks for');
+  ok(okSeg.prompt.includes('Sides: '), 'the skeleton locks the sides');
+  ok(okSeg.prompt.includes('Only @image'), 'the skeleton opens a single with who is in frame');
+  ok(okSeg.prompt.includes('looks toward screen-left'), 'the skeleton states the eyeline');
+  const noLook = framed(); delete noLook.seg.cuts[0].eyeline;
+  ok(only(failures('two-person-framing', p, noLook.ep)).some((m) => m.includes('is required')), 'an eyeline is required');
+  const vague = framed(); vague.seg.cuts[0].eyeline = { C01: 'at him' };
+  ok(only(failures('two-person-framing', p, vague.ep)).some((m) => m.includes('not valid')), '"at him" is refused');
+  const wrong = framed(); wrong.seg.cuts[0].eyeline = { C01: 'screen-right level' };
+  ok(only(failures('two-person-framing', p, wrong.ep)).some((m) => m.includes('so looking at the other person is screen-left')), 'looking away from the other person is caught');
+  const away = framed(); away.seg.cuts[0].eyeline = { C01: 'away screen-right' }; away.seg.prompt = promptSkeleton(p, away.ep, away.seg.id).video.prompt;
+  ok(!only(failures('two-person-framing', p, away.ep)).some((m) => m.includes('so looking')), 'a deliberate look away passes');
+  const noSides = framed(); noSides.seg.prompt = noSides.seg.prompt.replace(/Sides: [^\n]*\n/, '');
+  ok(only(failures('two-person-framing', p, noSides.ep)).some((m) => m.includes('must lock the sides')), 'a prompt without the sides line fails');
+  const leaked = framed();
+  const labelOf = /(@image\d+) looks toward screen-right/.exec(leaked.seg.prompt)[1];
+  leaked.seg.prompt = leaked.seg.prompt.replace(/(Shot 1 \([^)]*\): )/, `$1Behind him stands ${labelOf}. `).replace(/(\[Shot 1\] )/, `$1Behind him stands ${labelOf}. `);
+  ok(only(failures('two-person-framing', p, leaked.ep)).some((m) => m.includes('must not describe')), 'a single that names the other person fails');
+  const ots = framed(); const twoShot = ots.seg.cuts[2];
+  twoShot.ots = 'C01'; delete twoShot.eyeline.C01;
+  ots.seg.prompt = promptSkeleton(p, ots.ep, ots.seg.id).video.prompt;
+  ok(ots.seg.prompt.includes('their face not visible'), 'the skeleton writes the over-the-shoulder');
+  ok(ots.seg.prompt.includes('lower-right foreground'), 'the near shoulder sits on its own side of the axis');
+  eq(only(failures('two-person-framing', p, ots.ep)).length, 0, 'over-the-shoulder needs no eyeline from the shoulder');
+  const badOts = framed(); badOts.seg.cuts[0].ots = 'C02';
+  ok(only(failures('two-person-framing', p, badOts.ep)).some((m) => m.includes('ots must be one of')), 'ots must be on screen with one other');
+  eq(onlyInFramePhrase('@image3', 'zh'), '画面中只有@image3。', 'the Chinese single');
+  ok(overShoulderPhrase('@image3', 'left', 'zh').includes('左下方前景'), 'the Chinese over-the-shoulder');
+  ok(sidesLine({ screenLeft: 'A', screenRight: 'B' }, (id) => id, 'zh').includes('A始终在画面左侧'), 'the Chinese sides line');
+  // Keyframes carry the blocking in their pictures: the gate stays out of the way.
+  eq(failures('two-person-framing', PROJECT, E01).length, 0, 'keyframes mode is exempt');
+}
+
 {
   // The house cinematography (params.cinematography low-key): practical sources, no flat light on a
   // person, most cuts keyed from the side or behind, shot wide open; the look rides every prompt.
@@ -2307,8 +2416,8 @@ const breaks = (gateId, project, episode, needle) => {
 
 // ── gate inventory (crew engine) ────────────────────────────────────────────
 {
-  eq(GATES.length, 67, 'gate count');
-  eq(GATES.filter((g) => !g.advisory).length, 61, 'hard gate count');
+  eq(GATES.length, 69, 'gate count');
+  eq(GATES.filter((g) => !g.advisory).length, 63, 'hard gate count');
   eq(GATES.filter((g) => g.advisory).length, 6, 'advisory gate count');
   ok(!GATES.find((g) => g.id === 'boundary-contrast').advisory, 'boundary-contrast is hard');
   ok(GATES.find((g) => g.id === 'boundary-contrast-episode').advisory, 'boundary-contrast-episode is advisory');
@@ -2430,8 +2539,8 @@ const breaks = (gateId, project, episode, needle) => {
   const results = runGates(makeContext(PROJECT, { 1: E01 }));
   const rep = formatGates(results);
   eq(Object.keys(rep).sort().join(','), 'advisory,failed,hard,lines,notes,summary,text', 'formatGates shape');
-  eq(rep.summary, '✓ 61 gate(s) passed, 6 advisory (1 note)', 'fixture summary');
-  eq(rep.failed, 0, 'nothing failed'); eq(rep.hard, 61, 'hard count'); eq(rep.advisory, 6, 'advisory count');
+  eq(rep.summary, '✓ 63 gate(s) passed, 6 advisory (1 note)', 'fixture summary');
+  eq(rep.failed, 0, 'nothing failed'); eq(rep.hard, 63, 'hard count'); eq(rep.advisory, 6, 'advisory count');
   // The fixture now raises one advisory (E01-03 carries the 爽点 and is not cut as faceslap),
   // exactly as it does in the wji-cli copy — advisory lines print, failures still must not.
   eq(rep.lines.filter((l) => l.includes('✗')).length, 0, 'a green run prints no failed gates');