cinetic
코드에서 만든 시네마틱 런치 영상과 모션 디자인을 코딩 에이전트가 연출합니다
Kling 3.0, Veo 3, Wan, Nano Banana, Flux를 한 웹앱에서 AI 이미지와 영상을 만들 수 있다
Make AI images and videos with Kling 3.0, Veo 3, Wan, Nano Banana and Flux in one web app. See the cost before you generate; failed runs refund credits.
디자인PythonGitHub @uniqaiapp온라인운영자 미확인
써 보셨어요? 한 번만 눌러 주세요. 로그인은 필요 없어요.
Suno producer skill: write style prompts, meta-tags and lyric structure ([Verse], [Chorus], [Bridge]) for Suno v4.5+ by genre, mood and language, including Russian-language vocals. Claude Code skill + Cursor/Codex rules + a stdlib lint for tag placement and prompt length. Промпты для Suno на русском. By UniqAI.
README 전체를 불러오는 중입니다.
README 전문 보기 ↗Suno prompts, meta-tags and lyric structure that the model actually follows, in English and Russian.
Quickstart · How it works · Lint rules · Examples · Install · FAQ
English · Русский · 简体中文 · Español · Português (BR) · 日本語 · 한국어
Suno prompts in three parts: a one-line style prompt (genre, BPM, key, mood, instruments, production, vocal), a section structure with meta-tags ([Intro], [Verse], [Pre-Chorus], [Chorus], [Drop], [Bridge], [Outro]), and lyrics whose lines are even enough for a generated vocal to sing without rushing, written for the Style and Lyrics fields of the Suno editor. The tag list and field budgets were compiled against Suno v4.5 and later (v5.5 is the current model as of 2026-10, per Suno's March 2026 release coverage); the editor shows the live limits, and the lint budgets are flags you can override. It runs as a Claude Code skill, a Cursor rule or a Codex AGENTS.md snippet, and the three scripts work on their own as a plain checklist with a lint. UniqAI makes photo, video and chat tools, not music; this is a side project we use ourselves.
| Brief in, prompt out | Lyric meter, EN + RU | Lint before you spend credits |
|---|---|---|
brief.py turns "sad synthwave, female vocal" into a style prompt, two one-axis A/B variants, a tagged structure and per-section syllable targets. | lyric_meter.py counts syllables per line, derives the rhyme scheme per section and flags lines that drift from the section median. English heuristics, exact Russian vowel counting. | suno_lint.py checks tag spelling and placement, structure order, section length, field budgets and artist-imitation phrases. Stable rule ids, exit code 1 on errors. |
git clone https://github.com/uniqaiapp/suno-producer
cd suno-producer
python3 scripts/brief.py --genre synthwave --fuse "indie pop" --mood melancholic --seed 7
python3 scripts/lyric_meter.py examples/lyrics/midnight-bus.txt
python3 scripts/suno_lint.py examples/lyrics/bad-tags.txt --style examples/style/bad-style.txt
The first command prints a style prompt with variants B and C, a [Intro] … [Outro] structure and a lyric brief. The second prints syllables per line and the rhyme scheme of every section of the sample song. The third fails with three errors and nine warnings, each with a rule id and a line number. Python 3.9+, no packages to install.
Brief. scripts/brief.py reads references/genres.json (tempo range, keys, instruments, production words and default vocal per genre; editorial defaults, not measurements) and assembles a comma-separated style prompt in the order the model weights: genre, fusion, tempo, key, two mood words, instruments, production, vocal, language. With --fuse it uses the overlap of the two tempo ranges. It also picks the section structure for the genre and prints lines-per-section and syllable budgets.
Variants. Each variant changes exactly one descriptor (vocal, production, or tempo by about 8 %), so an A/B listen tells you which change mattered. tests/test_brief.py asserts the one-descriptor difference. Output is deterministic for a --seed.
Meter. scripts/lyric_meter.py splits the lyrics by [Section] tag, counts syllables (English: vowel groups with silent-e, -ed, -es and -le rules; Russian: one vowel, one syllable), builds the rhyme scheme from word endings (uppercase = full rhyme, lowercase = slant), and warns when a line is more than two syllables from the section median. Whole-line (backing vocals) are exempt.
Lint. scripts/suno_lint.py walks the lyrics once and reports against the rule table below, then checks the style prompt for tags, artist phrases, line count, tempo and budget. --json for machines, --strict to fail on warnings.
The skill (SKILL.md) ties it together for an agent: clarify the brief, run brief.py, write the lyrics into the structure, run both checks, deliver Style + Lyrics + variants in a fixed format, and when the user comes back with a bad generation, change one thing from references/troubleshooting.md.
description ──► brief.py ──► style prompt A/B/C + structure + lyric brief
│
write lyrics
│
lyric_meter.py ◄──┴──► suno_lint.py
syllables, scheme tags, order, budgets, L001–L016
│
Style + Lyrics + variants
여기까지가 README 앞부분입니다. 나머지는 원문에서 읽을 수 있습니다. README 전문 보기 ↗