MiniMax-AI/mcode-stack / stacks/jiachen/skills/making-content
making-content
UnverifiedContent production in three modes: rewrite, Xiaohongshu (小红书) image post, short video. Rewrite (default): turns an AI-sounding Chinese or mixed Chinese-English draft (post, newsletter, engineering blog, commentary) into the author's restrained style: few emoji, short paragraphs, hedged claims, neutral wording, English terms split by spaces. Triggers: 按我的风格改, 去 AI 味, 润色, 改改这篇, or a vague 帮我改改 when the context is a structured draft. Also for a Xiaohongshu image post from scratch (title, cover, cards, body, hooks); read references/xhs-post.md first. Also turns a document, research, notes or an article into a 60-180 s vertical video (抖音, 小红书, 视频号, B站: real sources, voice-over, captions, cover). Triggers: 做成视频, 做个短视频, 剪个 90 秒版本, 配音, 换成我的声音, or complaints of 字太密, AI 味太重, 容易走神; read references/video-pipeline.md first.
| File | Kind | Size |
|---|---|---|
| .gitignore | other | 167 B |
| SKILL.md | skill md | 25.6 KB |
| evals/evals.json | config | 4.5 KB |
| references/layouts.md | markdown | 8.5 KB |
| references/text-density-lint.md | markdown | 4.2 KB |
| references/video-pipeline.md | markdown | 12.5 KB |
| references/voice-and-music.md | markdown | 5.5 KB |
| references/xhs-post.md | markdown | 9.0 KB |
| requirements.txt | manifest | 233 B |
| scripts/capture.py | script | 6.4 KB |
| scripts/compositor.html | other | 27.8 KB |
| scripts/fetch_assets.py | script | 4.0 KB |
| scripts/lint.py | script | 8.6 KB |
| scripts/render.py | script | 4.6 KB |
| scripts/timeline.py | script | 8.7 KB |
| scripts/tts.py | script | 5.5 KB |
| templates/cover.html | other | 2.9 KB |
| templates/script.example.json | config | 5.0 KB |
| templates/sources.example.json | config | 58 B |
scripts/timeline.py
View on GitHub#!/usr/bin/env python3
"""Turn script.json + per-line TTS clips into timeline.json (shots, captions) and mix.wav.
python3 timeline.py --script script.json --tts tts --music music/Digital_Lemonade.mp3
* Shot cuts land on the word named by a shot's "anchor" (word timestamps from tts.py).
* Captions come from each line's "cap" ('/' and punctuation mark the breaks), re-split to one
short line each, and merged when a piece would flash by too fast. Punctuation is dropped.
* Audio = voice + music ducked under speech (whoosh on layout changes, off by default).
No per-element "pop" sounds (viewers found them distracting).
"""
import argparse, json, os, re, subprocess, sys
import numpy as np, soundfile as sf
LEAD, GAP, TAIL = 0.1, 0.1, 1.6 # s: silence before the first line / between lines / after the last
SR = 48000 # Hz; the rate render.py encodes AAC at
CAP_MAX = 16 # caption load per line (汉字 = 1, latin char = .5); Netflix zh-CN allows 16
CAP_MIN_DUR, CAP_MAX_CPS = 0.6, 9 # s on screen / chars per second (Netflix adult reading speed)
CUT_LEAD = 0.12 # s: cut slightly before the anchor word so the picture lands with the word
DUCK_SPEECH, DUCK_GAP = 0.11, 0.26 # music gain under speech / in pauses (~14 dB / ~6 dB below the voice)
DUCK_ON = 0.03 # envelope level that counts as speech (envelope normalised to 1)
WHOOSH_GAIN = 0.0 # transition whoosh off: viewers found sound effects odd; 0.18 was audible under the voice
def load(s):
return sum(1 if ord(ch) > 0x2e80 else 0.5 for ch in s if not ch.isspace() and ch not in ',。、:;?!,.:;?!“”"')
def keep(c):
return not (c.isspace() or c in ',、:;。?!,.:;?!')
def speakable(text): # must match tts.py
return re.sub(r'(?<=[A-Za-z0-9])[_-](?=[A-Za-z0-9])', ' ', text)
def split_cap(p):
if load(p) <= CAP_MAX:
return [p]
latin = lambda ch: ch.isascii() and (ch.isalnum() or ch in '_$%.+-')
cands = [i for i in range(1, len(p)) if not (latin(p[i - 1]) and latin(p[i]))
and not (p[i] == ' ' and i + 1 < len(p) and latin(p[i - 1]) and latin(p[i + 1])) and p[i] not in '的了吗呢']
half = load(p) / 2
i = min(cands, key=lambda i: abs(load(p[:i]) - half)) if cands else len(p) // 2
return split_cap(p[:i].strip()) + split_cap(p[i:].strip())
def merge_fast(cues, start):
"""Merge a too-short/too-fast caption into a neighbour, preferring not to cross a sentence end."""
bad = lambda c: c['e'] - c['s'] < CAP_MIN_DUR or load(c['text']) / max(c['e'] - c['s'], 0.01) > CAP_MAX_CPS
k = start
while k < len(cues):
if bad(cues[k]):
opts = []
for j in (k - 1, k + 1):
if start <= j < len(cues):
a, b = sorted((k, j))
m = {'s': cues[a]['s'], 'e': cues[b]['e'], 'text': cues[a]['text'] + ' ' + cues[b]['text'], 'end': cues[b]['end']}
if load(m['text']) <= CAP_MAX:
opts.append(((10 if cues[a]['end'] else 0) + load(m['text']) / (m['e'] - m['s']), a, b, m))
if opts:
_, a, b, m = min(opts, key=lambda o: o[0]); cues[a] = m; del cues[b]; k = start; continue
k += 1
def duration(path):
out = subprocess.run(['ffprobe', '-v', 'error', '-show_entries', 'format=duration', '-of', 'csv=p=0', path], capture_output=True, text=True).stdout
if not out.strip():
sys.exit(f'cannot read {path}; run tts.py first')
return float(out)
def build(script, tts_dir):
t, lines, shots, cues = LEAD, [], [], []
for i, x in enumerate(script['lines']):
base = os.path.join(tts_dir, f'{i:02d}')
d = duration(base + '.mp3'); j = json.load(open(base + '.json'))
text, T = j['text'], []
for w in j['words']:
cs = [c for c in w['word'] if keep(c)]
T += [w['startTime'] + (w['endTime'] - w['startTime']) * k / max(1, len(cs)) for k in range(len(cs))]
T.append(j['words'][-1]['endTime'] if j['words'] else d)
def at(anchor):
a = speakable(anchor)
if a not in text:
sys.exit(f'line {i}: anchor {anchor!r} not found in narration')
return t + T[min(sum(1 for c in text[:text.index(a)] if keep(c)), len(T) - 1)]
lines.append({'i': i, 'start': round(t, 3), 'dur': d})
for k, sh in enumerate(x['shots']):
st = 0 if (i == 0 and k == 0) else (at(sh['anchor']) - CUT_LEAD if sh.get('anchor') else t)
sh = {**sh, 'start': round(max(0, st), 3), 'chapter': x['chapter']}
if sh['type'] == 'stack':
sh['items'] = [{**it, 'at': round(at(it['anchor']), 3)} for it in sh['items']]
shots.append(sh)
pieces, ends = [], []
for r in re.findall(r'[^,、:;。?!/]+[,、:;。?!/]?', x['cap']):
body = r.rstrip(',、:;。?!/').strip()
if body:
sub = split_cap(body); pieces += sub; ends += [False] * (len(sub) - 1) + [r[-1:] in '。?!']
tot, acc, n_, first = sum(load(p) for p in pieces) or 1, 0, len(T) - 1, len(cues)
for p, e in zip(pieces, ends):
k0 = round(acc / tot * n_); acc += load(p); k1 = round(acc / tot * n_)
cues.append({'s': round(t + T[k0], 3), 'e': round(t + T[min(k1, n_)], 3), 'text': p, 'end': e})
merge_fast(cues, first)
t += d + GAP
total = round(t - GAP + TAIL, 3)
for k, sh in enumerate(shots):
sh['end'] = shots[k + 1]['start'] if k + 1 < len(shots) else total
return lines, shots, cues, total
def whoosh(rng):
n = int(0.42 * SR); x = rng.standard_normal(n); out = np.zeros(n); hop = 512
for s0 in range(0, n, hop): # band-pass noise with a rising centre frequency
seg = x[s0:s0 + hop * 2]; f = np.fft.rfft(seg, n=hop * 2); fr = np.fft.rfftfreq(hop * 2, 1 / SR)
c = 300 + 5000 * (s0 / n); f *= np.exp(-((fr - c) / (c * 0.6)) ** 2); y = np.fft.irfft(f)[:len(seg)]
out[s0:s0 + len(y)] += y * np.hanning(len(y))
return out / np.abs(out).max() * np.sin(np.pi * np.linspace(0, 1, n)) ** 1.5 * WHOOSH_GAIN
def mix(lines, shots, total, tts_dir, music, out, tmp):
n = int(total * SR)
ins = sum([['-i', os.path.join(tts_dir, f"{L['i']:02d}.mp3")] for L in lines], [])
flt = ';'.join(f"[{k}:a]aresample={SR},adelay={int(L['start'] * 1000)}|{int(L['start'] * 1000)}[a{k}]" for k, L in enumerate(lines))
flt += ';' + ''.join(f'[a{k}]' for k in range(len(lines))) + f'amix=inputs={len(lines)}:normalize=0,apad,atrim=0:{total},pan=mono|c0=c0[o]'
vpath = os.path.join(tmp, 'voice.wav')
subprocess.run(['ffmpeg', '-y', '-v', 'error', *ins, '-filter_complex', flt, '-map', '[o]', '-ar', str(SR), '-ac', '1', vpath], check=True)
voice, _ = sf.read(vpath); voice = np.pad(voice, (0, max(0, n - len(voice))))[:n]
bed = np.zeros((n, 2))
if music:
mpath = os.path.join(tmp, 'music.wav')
subprocess.run(['ffmpeg', '-y', '-v', 'error', '-i', music, '-ar', str(SR), '-ac', '2', mpath], check=True)
m, _ = sf.read(mpath); m = np.tile(m, (int(np.ceil(n / len(m))), 1))[:n]
env = np.convolve(np.abs(voice), np.ones(int(0.25 * SR)) / int(0.25 * SR), mode='same'); env /= env.max() + 1e-9
duck = np.convolve(np.where(env > DUCK_ON, DUCK_SPEECH, DUCK_GAP), np.ones(int(0.3 * SR)) / int(0.3 * SR), mode='same')
fade = np.ones(n); fade[-int(2.5 * SR):] = np.linspace(1, 0, int(2.5 * SR)); fade[:int(0.4 * SR)] = np.linspace(0, 1, int(0.4 * SR))
bed = m * (duck * fade)[:, None]
fx, rng = np.zeros(n), np.random.default_rng(7)
for k in range(1, len(shots)):
if shots[k]['type'] != shots[k - 1]['type']:
w = whoosh(rng); i0 = max(0, int((shots[k]['start'] - CUT_LEAD) * SR)); seg = w[:max(0, n - i0)]; fx[i0:i0 + len(seg)] += seg
m = bed + (voice + fx)[:, None]
sf.write(out, m / max(1.0, np.abs(m).max() / 0.97), SR)
def main():
ap = argparse.ArgumentParser()
ap.add_argument('--script', default='script.json'); ap.add_argument('--tts', default='tts')
ap.add_argument('--music', help='background track (looped, ducked under the voice)')
ap.add_argument('--out', default='timeline.json'); ap.add_argument('--mix', default='mix.wav'); ap.add_argument('--tmp', default='.')
a = ap.parse_args()
script = json.load(open(a.script))
lines, shots, cues, total = build(script, a.tts)
json.dump({'meta': script.get('meta', {}), 'shots': shots, 'cues': cues, 'total': total, 'lines': lines},
open(a.out, 'w'), ensure_ascii=False)
mix(lines, shots, total, a.tts, a.music, a.mix, a.tmp)
print(f'{total:.1f}s, {len(shots)} shots, {len(cues)} captions -> {a.out}, {a.mix}')
if __name__ == '__main__':
main()
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| xlsxanthropics/skillsUse this skill any time a spreadsheet file is the primary input or output. This means any task where the user wants to: open, read, edit, or fix an existing .xlsx, .xlsm, .xltx, .csv, or .tsv file (e.g., adding columns, computing formulas, formatting, charting, cleaning messy data); create a new spreadsheet from scratch or from other data sources; or convert between tabular file formats. Trigger especially when the user references a spreadsheet file by name or path — even casually (like "the xlsx in my downloads") — and wants something done to it or produced from it. Also trigger for cleaning or restructuring messy tabular data files (malformed rows, misplaced headers, junk data) into proper spreadsheets. The deliverable must be a spreadsheet file. Do NOT trigger when the primary deliverable is a Word document, HTML report, standalone Python script, database pipeline, or Google Sheets API integration, even if tabular data is involved. | Stars: 195k | Updated | Claude CodeCodex | Verified | Install |
| shipping-and-launchaddyosmani/agent-skillsPrepares production launches. Use when preparing to deploy to production, or when asking what needs to be in place before shipping. Use when you need a pre-launch checklist, when setting up monitoring, when planning a staged rollout, or when you need a rollback strategy. | Stars: 118k | Updated | Claude CodeCodex | Verified | Install |
| childhoodthedotmack/claude-memBabysit a shipped release after merge, publication, or deployment. Verify published packages, production behavior, installed plugins, and late CI or user reports until the observation window is clean; use babysit for PR readiness before release. | Stars: 117k | Updated | Claude CodeCodex | Unverified | Install |
| paperclip-pagepaperclipai/paperclipPublish static HTML pages and asset folders to the Paperclip S3/CloudFront page host. Use when asked to deploy, host, or share a persistent page, viewer, prototype, report, or static site without here.now. | Stars: 114k | Updated | Claude CodeCodex | Verified | Install |