Ancient-Poetic Chinese Lyric Prompts: 10 词牌 / 诗化 Templates

Not allusion-stacked fake-ancient — genuinely poeticized, structured ancient-Chinese lyrics. Ten templates with explicit character count, rhyme, imagery and allusion-density caps.

Ancient-poetic (古风诗意) lyrics sit closer to classical 词 / 赋 density than generic “Chinese-style” pop: parallelism, tonal pattern and allusion all matter, but density is capped at most one obscure word per verse. The ten prompts below pin down character count, rhyme, an imagery whitelist and allusion density, so the model produces something singable instead of a wall of stacked references. Each is ready for ancient-style originals, wuxia / xianxia themes, and recitation pieces.

TL;DR

  • A strong ancient-poetic prompt names six things: character count, rhyme, imagery whitelist, allusion cap, a banned-word list, and one explicit emotion.
  • The fatal failure mode is allusion overload. Force one concrete object plus one action per verse and the “fake-ancient” feel disappears.
  • For the writing step, Claude Opus 4.7 / Sonnet 4.6 give the cleanest cadence; a Chinese-trained model like DeepSeek often handles classical idiom and 平仄 better. Draft, then pick the line you like.
  • For the singing step, paste Chinese characters (not pinyin) into Suno v5.5 Custom Lyrics; keep lyrics under 3,000 characters or Suno silently truncates.

What a high-quality prompt should contain

Six required elements:

  • Explicit character count: 7 (七言) / 5 (五言) / 5–7 alternating (词牌-feel)
  • Rhyme spec: ang / ong / eng / ou — a consistent final reads “ancient”
  • Imagery whitelist: name only the allowed images (river, moon, sword, wine, horse / lamp, candle, curtain, night)
  • Allusion density: at most one per verse — more becomes stacking
  • Banned list: 红尘 / 苍生 / 万古 / eternity — the markers of fake-ancient
  • Explicit emotion: clear-eyed reminiscence / detached calm / heavy but not weeping / resolve under shock

10 copy-ready prompt templates

1. 7-character long song

Best for: Original ancient-Chinese single

写一首七言古风长歌词,主题:江湖侠客中年回顾少年志。要求:每句严格 7 字;4 段主歌 + 2 段副歌;偶句押 ang 韵;每段最多 1 个生僻词;用江、月、剑、酒、马等具体意象;避免堆砌"红尘 / 苍生 / 万古"。

2. 词牌-style mixed line

Best for: Ancient female-vocal single

写一首长短句古风词牌感歌词,主题:闺中思人。结构按 5/7/5/7/7 字一段循环;押 ou / ong 韵;意象只用:灯、烛、帘、夜、月、楼;每段必须含 1 个具体动作(吹灯 / 卷帘 / 提笔 / 折花);不可超过 1 个典故。

3. Youthful sword-wandering

Best for: Wuxia drama theme

写一首中文古风诗意歌词,主题:少年仗剑天涯。结构:主歌 1 / 副歌 / 主歌 2 / 副歌 / Bridge / 副歌。每段必须含一个具体地点(如:长安、洛阳、玉门关、燕然)+ 一个具体动作;押 an / uan 韵;副歌出现一句"少年"开头的核心句,每次出现微调一个字。

4. Palace late-years memory

Best for: Period palace drama theme

写一首古风女声歌词,主题:宫中后妃晚年回忆少年入宫。第一人称视角。结构按 7/5/7/5 字段循环;押 ai 韵;意象用:金殿、玉阶、灯、镜、纱;情绪是"清醒回望",不是哭;每段最多 1 个典故。

5. Immortal-mortal love

Best for: Xianxia drama, Chinese animation theme

写一首古风仙侠歌词,主题:神仙下凡爱上凡人。第三人称叙事。结构:主歌 / 副歌 / 主歌 / Bridge / 副歌。每段含一个仙凡对比意象(云 vs 尘、永恒 vs 一夕);副歌出现"凡心" / "仙骨" / "尘世"等核心词但每段微调;押 ong 韵。

6. Frontier soldier homesick

Best for: War-history theme

写一首古风战争主题歌词,主题:戍边将士冬日思乡。第一人称男声视角。每句 7 字;4 段主歌 + 2 段副歌;押 ang 韵;意象用:边、雪、剑、马、月、家书;情绪是"沉重但不哭";副歌出现"望故乡"。

7. Farewell song

Best for: Period drama interlude

写一首古风女子离别歌词,主题:青楼女子送恩客远行。第一人称女声视角。结构 7/5 字段交替;押 ou 韵;意象用:扇、酒、灯、夜、巷、马蹄;不写"再见"二字;副歌通过物件传情;不可超过 2 个典故。

8. Hermit autumn-mountain

Best for: Scholar aesthetic, Chinese brand

写一首古风诗化山水主题歌词,主题:隐士秋日观山。第一人称视角。结构每段 5 字 + 7 字交替;押 eng / ing 韵;意象用:松、霜、菊、酒、孤鹤;情绪是"清淡寡欲";每段最多 1 个动作;不可有大段抒情,只写所见所闻。

9. Frontier woman

Best for: Wuxia / historical drama

写一首古风边塞女子主题歌词,主题:将军之女初到边塞。第一人称视角。每句 7 字;3 段主歌 + 副歌反复;押 uan 韵;意象用:沙、马、剑、长城、北风;情绪是"惊与决";副歌出现"塞外"。

10. Wei bridge willow farewell

Best for: Ancient love single

写一首古风春日离别歌词,主题:恋人在长安渭桥折柳送别。第一人称视角。结构:主歌 / 副歌 / 主歌 / Bridge / 副歌。每句 5 字;押 i 韵;意象只用:柳、桥、风、酒、马、春雨;不写"再见";副歌反复"渭桥";典故只用一次(长安灞桥折柳即可)。

Which template for which goal

You wantUse templateKey spec
Strict 7-character long song1every line 7 chars, ang rhyme
词牌 (mixed-line) feel2 / 105–7 alternating, ou / ong / i rhyme
Narrative ancient3 / 5 / 9verse-chorus + a named place each verse
Palace, first-person4gold throne / jade steps, ai rhyme
Frontier / war6 / 9sand, sword, horse, Great Wall
Hermit / scholar quiet8sees-and-hears only, no broad emoting

Which model to write with (June 2026)

You draft the lyrics in a chat model, then sing them in Suno. The two steps reward different tools.

ModelStrength hereNote
Claude Opus 4.7 / Sonnet 4.6Cleanest cadence and line discipline; follows the character-count rule wellSonnet 4.6 is the cheaper workhorse; Opus 4.7 for the final polish
DeepSeek (Chinese-trained)Best feel for classical idiom and 平仄; more allusions on tapStrongest at avoiding “translated-sounding” Chinese
GPT-5.5Solid all-rounder; reliable structurePicker: use Thinking mode for the count-and-rhyme constraints
Gemini 3.1 ProGood long-context drafting of multiple variants at once1M context lets you keep the whole brief plus past drafts

Practical move: run the same prompt through two models, then keep the strongest verse from each. Constraint-following (exact character count, single rhyme) is where these models still slip, so reading the line aloud beats trusting the first pass.

Common mistakes

  • Allusion overload — the fatal trap. One per verse, no more.
  • Mixing 诗 / 词 / 赋 formats — pick one per song.
  • Obscure-character overload — “ancient” is not “illegible.”
  • Empty lyricism (thousand-year passion) — replace with a concrete object plus an action.
  • No character-count spec — the model writes uneven lines and Suno’s phrasing wobbles.

Taking it to Suno v5.5

Once the lyrics read well on the page, the singing step has its own rules (as of June 2026):

  • Paste Chinese characters, not pinyin, into Custom Lyrics mode. v5.5 improved Chinese and dialect articulation, but it still pronounces characters far more reliably than romanized text.
  • Keep total lyrics under 3,000 characters. Suno truncates past that silently, so a long 七言 piece can lose its final chorus without warning.
  • Put the instrumentation in the style field (up to ~1,000 characters on v5 / v5.5), e.g. guzheng, dizi, light modern drums, female vocal, cinematic.
  • Generate two or three takes and pick the cleanest pronunciation. Ancient diction (生僻字, place names like 燕然) is where Suno is most likely to mangle a syllable.

FAQ

Q: How do I avoid the “edgy fake-ancient” feel?

A: Ban the big words (万古 / 苍生 / 永恒) and require one concrete object plus one action per verse. Breaking a willow branch at the bridge lands harder than eternal love every time.

Q: Do I have to write in classical Chinese?

A: No. Ancient feel is not the same as classical grammar. Writing an ancient mood in modern Chinese is the higher path — think Vincent Fang’s lyrics for Jay Chou.

Q: Can I just paste in a real 词牌 like 《忆江南》?

A: Use it for the character count only. As of June 2026, models still wobble on exact tonal patterns (平仄). Specify count plus rhyme plus imagery and you get a convincing 词牌-feel without fighting the metre.

Q: Which AI writes the best ancient-poetic lyrics?

A: For clean cadence and rule-following, Claude Opus 4.7 / Sonnet 4.6. For the most idiomatic classical flavor, a Chinese-trained model such as DeepSeek. Drafting through both and keeping the best lines is faster than perfecting one prompt.

Q: Can Suno sing Chinese ancient vocals well?

A: Better than before. Suno v5.5 sharpened Chinese and dialect articulation, but obscure characters still trip it up. Paste characters in Custom Lyrics, generate a few takes, or render an instrumental and record a real vocal over it.

Q: What music pairs with ancient-poetic lyrics?

A: Chinese fusion (guzheng plus modern drums) or pure classical (guzheng, dizi, bianzhong). See the Chinese fusion guide below for ready style prompts.

Tags: #Lyrics #Ancient poetic #Chinese-style #Chinese #Prompt