song-generation

歌曲生成

使用 each::sense 生成包含人声和完整器乐制作的完整歌曲——这是一款智能 AI Agent，可自动为你的请求选择最佳模型。

快速开始

需要 each::labs API 密钥。请前往 eachlabs.ai 获取。

使用 curl

curl -X POST https://eachsense-agent.core.eachlabs.run/v1/chat/completions \
  -H "Content-Type: application/json" \
  -H "X-API-Key: $EACHLABS_API_KEY" \
  -d '{
    "messages": [{"role": "user", "content": "Generate an upbeat indie pop song about road trips and summer freedom, female vocal, 120 BPM, with acoustic guitar, claps, and a catchy whistling hook in the chorus"}],
    "stream": false
  }'

使用 Python（OpenAI SDK）

from openai import OpenAI

client = OpenAI(
    api_key="YOUR_EACHLABS_API_KEY",
    base_url="https://eachsense-agent.core.eachlabs.run/v1"
)

response = client.chat.completions.create(
    model="eachsense/beta",
    messages=[{"role": "user", "content": "Generate an upbeat indie pop song about road trips and summer freedom, female vocal, 120 BPM, with acoustic guitar, claps, and a catchy whistling hook in the chorus"}]
)

print(response.choices[0].message.content)

使用参考图像

curl -X POST https://eachsense-agent.core.eachlabs.run/v1/chat/completions \
  -H "Content-Type: application/json" \
  -H "X-API-Key: $EACHLABS_API_KEY" \
  -d '{
    "messages": [{"role": "user", "content": [
              {"type": "text", "text": "Generate a song inspired by the mood and atmosphere of this image. Match the emotion, whether it feels melancholy, joyful, or mysterious."},
              {"type": "image_url", "image_url": {"url": "https://example.com/mood-photo.jpg"}}
            ]
          }
    ],
    "stream": false
  }'

图像通过 OpenAI 多模态内容格式放在消息中发送。每个请求最多可包含 4 张图像。

流式传输

设置 "stream": true 可获取实时 SSE 响应，设置 `"stream": false" 可在单个响应中获取完整结果。流式传输适用于在 UI 中显示进度；非流式传输则更适合脚本和自动化。

可用模型

模型	优势	最适合
mureka-generate-music	生成带有人声、歌词和制作的完整歌曲	各种流派的完整歌曲
stable-audio-2-5	高质量短音频生成	短音乐片段、抓耳段落、前奏

each::sense 会自动选择合适的模型。对于带人声的完整歌曲，主要使用 mureka-generate-music。

歌曲结构指南

常见歌曲结构

结构	模式	时长
流行标准结构	主歌 - 副歌 - 主歌 - 副歌 - 桥段 - 副歌	3-4 分钟
简单结构	主歌 - 副歌 - 主歌 - 副歌	2-3 分钟
扩展结构	前奏 - 主歌 - 预副歌 - 副歌 - 主歌 - 预副歌 - 副歌 - 桥段 - 副歌 - 尾奏	4-5 分钟
嘻哈	前奏 - 主歌 - Hook - 主歌 - Hook - 主歌 - Hook - 尾奏	3-4 分钟
抒情曲	前奏 - 主歌 - 主歌 - 副歌 - 主歌 - 副歌 - 尾奏	3-5 分钟

提供歌词

在提示词中直接加入歌词，并使用结构标记：

Generate a pop rock song, male vocal, 110 BPM:

[Verse 1]
Walking down these empty streets at dawn
Shadows fading as the night moves on
Every step I take leads back to you
Underneath this sky of fading blue

[Chorus]
We were golden, we were fire
Burning brighter, climbing higher
Now I'm standing in the rain
Chasing echoes of your name

流派指南

流派	常见 BPM	主要描述词	人声风格
流行	100-130	抓耳的 Hook、精致的制作	清澈、明亮、有和声
摇滚	110-140	电吉他、强劲的鼓点	有力、粗犷、富有变化
嘻哈	80-100	808 低音、踩镲、Trap 节拍	有节奏感的流动感、说唱
R&B	60-90	柔和的低音、丰富的和弦	深情、富有旋律变化
乡村	90-130	原声吉他、小提琴、钢棒吉他	温暖、叙事性强
电子	120-150	合成器、Drop、渐强段落	经过处理、Vocoded、切片
独立	100-130	Lo-fi 质感、清脆的吉他	亲密、对话式
爵士	80-140	摇摆、即兴、铜管乐器	流畅、拟声哼唱

提示词技巧

人声设定

明确描述人声特征：

"female vocal, alto range, smooth and breathy tone"
"male vocal, deep baritone, raspy rock voice"
"duet with male and female vocals trading verses"
"choir-style harmonies in the chorus"

氛围与情绪

叠加情绪描述词，生成细腻丰富的效果：

"bittersweet and nostalgic, like remembering a perfect day you can never return to"
"triumphant and empowering, building from quiet determination to explosive confidence"

乐器细节

列出具体乐器以便精准控制：

"acoustic guitar, brush drums, upright bass, warm Rhodes piano, subtle strings"
"heavy distorted electric guitar, double kick drums, bass guitar, screaming lead solo"

示例

自定义歌词的流行歌曲

curl -X POST https://eachsense-agent.core.eachlabs.run/v1/chat/completions \
  -H "Content-Type: application/json" \
  -H "X-API-Key: $EACHLABS_API_KEY" \
  -d '{
    "messages": [{"role": "user", "content": "Generate an upbeat pop song, female vocal, 118 BPM, with synth bass and claps:\n\n[Verse 1]\nWoke up to the golden light\nEverything is feeling right\nGot my coffee, got my song\nNothing here can go wrong\n\n[Chorus]\nThis is our summer, this is our time\nDancing in the sunshine, feeling so alive\nWe are unstoppable, we are on fire\nTaking it higher and higher"}],
    "stream": false
  }'

摇滚颂歌

curl -X POST https://eachsense-agent.core.eachlabs.run/v1/chat/completions \
  -H "Content-Type: application/json" \
  -H "X-API-Key: $EACHLABS_API_KEY" \
  -d '{
    "messages": [{"role": "user", "content": "Generate an arena rock anthem about never giving up. Male vocal with a gritty, powerful voice. 130 BPM. Driving electric guitar riff, pounding drums, bass guitar, and a massive singalong chorus. Think Foo Fighters meets Queen. 3 minutes long."}],
    "stream": false
  }'

嘻哈曲目

curl -X POST https://eachsense-agent.core.eachlabs.run/v1/chat/completions \
  -H "Content-Type: application/json" \
  -H "X-API-Key: $EACHLABS_API_KEY" \
  -d '{
    "messages": [{"role": "user", "content": "Generate a chill hip-hop track about city nights and ambition. Male rap vocal with a smooth flow. 88 BPM. Dark trap beat with 808 bass, hi-hats, ambient synth pads, and a soulful R&B female vocal hook on the chorus. Moody and atmospheric."}],
    "stream": false
  }'

原声抒情曲

curl -X POST https://eachsense-agent.core.eachlabs.run/v1/chat/completions \
  -H "Content-Type: application/json" \
  -H "X-API-Key: $EACHLABS_API_KEY" \
  -d '{
    "messages": [{"role": "user", "content": "Generate a tender acoustic ballad about saying goodbye to a childhood home. Female vocal, intimate and emotional. 72 BPM. Fingerpicked acoustic guitar, soft cello, and gentle piano. Build to a fuller arrangement in the final chorus with strings. Bittersweet and moving."}],
    "stream": false
  }'

广告歌 / 短歌曲

curl -X POST https://eachsense-agent.core.eachlabs.run/v1/chat/completions \
  -H "Content-Type: application/json" \
  -H "X-API-Key: $EACHLABS_API_KEY" \
  -d '{
    "messages": [{"role": "user", "content": "Generate a 15-second catchy jingle for a coffee brand called Morning Brew. Upbeat and cheerful, male and female vocals harmonizing, ukulele and handclaps, with the tagline: Rise and brew, the best in you. Earworm melody that sticks."}],
    "stream": false
  }'

常见问题

• 没有人声指导会生成千篇一律的人声。请务必指定性别、风格和能量水平。
• 未指定速度会让模型自行猜测 BPM，可能与你的设想不符。请指定 BPM。
• 歌词没有结构标记会让模型难以正确区分主歌和副歌。
• 混合过多流派（例如“爵士-金属-乡村-电子融合”）会削弱效果。最多融合 2 种流派。
• 忽略歌曲时长可能会生成过短或过长的片段，不符合你的需求。

相关技能

• 歌词生成 — 先生成歌词，再制作歌曲
• 音乐生成 — 不带人声的纯器乐音轨
• 语音生成 — 生成口语或旁白音频
• 音效 — 为作品叠加音效

文档

• each::sense 概览
• each::labs API