Ace-Step-Munk / docs /zh /Openrouter_API_DOC.md
OnyxMunk's picture
Add LoRA training assets: scripts, docs (no binaries), ui, my_dataset
bc9c638
# ACE-Step OpenRouter API 文档
> 兼容 OpenAI Chat Completions 格式的 AI 音乐生成接口
**Base URL:** `http://{host}:{port}` (默认 `http://127.0.0.1:8002`)
---
## 目录
- [认证](#认证)
- [接口列表](#接口列表)
- [POST /v1/chat/completions - 生成音乐](#1-生成音乐)
- [GET /v1/models - 模型列表](#2-模型列表)
- [GET /health - 健康检查](#3-健康检查)
- [输入模式](#输入模式)
- [音频输入](#音频输入)
- [流式响应](#流式响应)
- [完整示例](#完整示例)
- [错误码](#错误码)
---
## 认证
如果服务端配置了 API Key(环境变量 `OPENROUTER_API_KEY` 或启动参数 `--api-key`),所有请求需在 Header 中携带:
```
Authorization: Bearer <your-api-key>
```
未配置 API Key 时无需认证。
---
## 接口列表
### 1. 生成音乐
**POST** `/v1/chat/completions`
通过聊天消息生成音乐,返回音频数据和 LM 生成的元信息。
#### 请求参数
| 字段 | 类型 | 必填 | 默认值 | 说明 |
|---|---|---|---|---|
| `model` | string | 否 | 自动 | 模型 ID(从 `/v1/models` 获取) |
| `messages` | array | **是** | - | 聊天消息列表,见 [输入模式](#输入模式) |
| `stream` | boolean | 否 | `false` | 是否启用流式返回,见 [流式响应](#流式响应) |
| `audio_config` | object | 否 | `null` | 音频生成配置,见下方 |
| `temperature` | float | 否 | `0.85` | LM 采样温度 |
| `top_p` | float | 否 | `0.9` | LM nucleus sampling |
| `seed` | int \| string | 否 | `null` | 随机种子。`batch_size > 1` 时可用逗号分隔指定多个,如 `"42,123,456"` |
| `lyrics` | string | 否 | `""` | 直接传入歌词(优先级高于 messages 中解析的歌词),此时 messages 文本作为 prompt |
| `sample_mode` | boolean | 否 | `false` | 启用 LLM sample 模式,messages 文本作为 sample_query 由 LLM 自动生成 prompt/lyrics |
| `thinking` | boolean | 否 | `false` | 是否启用 LLM thinking 模式(更深度推理) |
| `use_format` | boolean | 否 | `false` | 当用户提供 prompt/lyrics 时,是否先通过 LLM 格式化增强 |
| `use_cot_caption` | boolean | 否 | `true` | 是否通过 CoT 改写/增强音乐描述 |
| `use_cot_language` | boolean | 否 | `true` | 是否通过 CoT 自动检测歌词语言 |
| `guidance_scale` | float | 否 | `7.0` | Classifier-free guidance scale |
| `batch_size` | int | 否 | `1` | 生成音频数量 |
| `task_type` | string | 否 | `"text2music"` | 任务类型,见 [音频输入](#音频输入) |
| `repainting_start` | float | 否 | `0.0` | repaint 区域起始位置(秒) |
| `repainting_end` | float | 否 | `null` | repaint 区域结束位置(秒) |
| `audio_cover_strength` | float | 否 | `1.0` | cover 强度 (0.0~1.0) |
#### audio_config 对象
| 字段 | 类型 | 默认值 | 说明 |
|---|---|---|---|
| `duration` | float | `null` | 音频时长(秒),不传由 LM 自动决定 |
| `bpm` | integer | `null` | 每分钟节拍数,不传由 LM 自动决定 |
| `vocal_language` | string | `"en"` | 歌词语言代码(如 `"zh"`, `"en"`, `"ja"`) |
| `instrumental` | boolean | `null` | 是否为纯器乐(无人声)。不传时根据歌词自动判断 |
| `format` | string | `"mp3"` | 输出音频格式 |
| `key_scale` | string | `null` | 调号(如 `"C major"`) |
| `time_signature` | string | `null` | 拍号(如 `"4/4"`) |
> **messages 文本含义取决于模式:**
> - 设置了 `lyrics` → messages 文本 = prompt(音乐描述)
> - 设置了 `sample_mode: true` → messages 文本 = sample_query(交给 LLM 生成一切)
> - 均未设置 → 自动检测:有标签走标签模式,像歌词走歌词模式,否则走 sample 模式
#### messages 格式
支持纯文本和多模态(文本 + 音频)两种格式:
**纯文本:**
```json
{
"messages": [
{"role": "user", "content": "你的输入内容"}
]
}
```
**多模态(含音频输入):**
```json
{
"messages": [
{
"role": "user",
"content": [
{"type": "text", "text": "翻唱这首歌"},
{
"type": "input_audio",
"input_audio": {
"data": "<base64 音频数据>",
"format": "mp3"
}
}
]
}
]
}
```
---
#### 非流式响应 (`stream: false`)
```json
{
"id": "chatcmpl-a1b2c3d4e5f6g7h8",
"object": "chat.completion",
"created": 1706688000,
"model": "acemusic/acestep-v15-turbo",
"choices": [
{
"index": 0,
"message": {
"role": "assistant",
"content": "## Metadata\n**Caption:** Upbeat pop song...\n**BPM:** 120\n**Duration:** 30s\n**Key:** C major\n\n## Lyrics\n[Verse 1]\nHello world...",
"audio": [
{
"type": "audio_url",
"audio_url": {
"url": "data:audio/mpeg;base64,SUQzBAAAAAAAI1RTU0UAAAA..."
}
}
]
},
"finish_reason": "stop"
}
],
"usage": {
"prompt_tokens": 10,
"completion_tokens": 100,
"total_tokens": 110
}
}
```
**响应字段说明:**
| 字段 | 说明 |
|---|---|
| `choices[0].message.content` | LM 生成的文本信息,包含 Metadata(Caption/BPM/Duration/Key/Time Signature/Language)和 Lyrics。如果 LM 未参与,返回 `"Music generated successfully."` |
| `choices[0].message.audio` | 音频数据数组,每项包含 `type` (`"audio_url"`) 和 `audio_url.url`(Base64 Data URL,格式 `data:audio/mpeg;base64,...`) |
| `choices[0].finish_reason` | `"stop"` 表示正常完成 |
**音频解码格式:**
`audio_url.url` 值为 Data URL 格式:`data:audio/mpeg;base64,<base64_data>`
客户端提取 base64 数据部分后解码即可得到 MP3 文件:
```python
import base64
url = response["choices"][0]["message"]["audio"][0]["audio_url"]["url"]
# 去掉 "data:audio/mpeg;base64," 前缀
b64_data = url.split(",", 1)[1]
audio_bytes = base64.b64decode(b64_data)
with open("output.mp3", "wb") as f:
f.write(audio_bytes)
```
```javascript
const url = response.choices[0].message.audio[0].audio_url.url;
const b64Data = url.split(",")[1];
const audioBytes = atob(b64Data);
// 或直接用于 <audio> 标签
const audio = new Audio(url);
audio.play();
```
---
### 2. 模型列表
**GET** `/v1/models`
返回可用模型信息。
#### 响应
```json
{
"data": [
{
"id": "acemusic/acestep-v15-turbo",
"name": "ACE-Step",
"created": 1706688000,
"description": "High-performance text-to-music generation model. Supports multiple styles, lyrics input, and various audio durations.",
"input_modalities": ["text", "audio"],
"output_modalities": ["audio", "text"],
"context_length": 4096,
"pricing": {"prompt": "0", "completion": "0", "request": "0"},
"supported_sampling_parameters": ["temperature", "top_p"]
}
]
}
```
---
### 3. 健康检查
**GET** `/health`
#### 响应
```json
{
"status": "ok",
"service": "ACE-Step OpenRouter API",
"version": "1.0"
}
```
---
## 输入模式
系统根据 `messages` 中最后一条 `user` 消息的内容自动选择输入模式。也可通过 `lyrics``sample_mode` 字段显式指定。
### 模式 1: 标签模式(推荐)
使用 `<prompt>``<lyrics>` 标签明确指定音乐描述和歌词:
```json
{
"messages": [
{
"role": "user",
"content": "<prompt>A gentle acoustic ballad in C major, female vocal</prompt>\n<lyrics>[Verse 1]\nSunlight through the window\nA brand new day begins\n\n[Chorus]\nWe are the dreamers\nWe are the light</lyrics>"
}
],
"audio_config": {
"duration": 30,
"vocal_language": "en"
}
}
```
- `<prompt>...</prompt>` — 音乐风格/场景描述(即 caption)
- `<lyrics>...</lyrics>` — 歌词内容
- 两个标签可以只传其中一个
-`use_format: true` 时,LLM 会自动增强 prompt 和 lyrics
### 模式 2: 自然语言模式(Sample 模式)
直接用自然语言描述想要的音乐,系统自动通过 LLM 生成 prompt 和 lyrics:
```json
{
"messages": [
{"role": "user", "content": "帮我生成一首欢快的中文流行歌曲,关于夏天和旅行"}
],
"sample_mode": true,
"audio_config": {
"vocal_language": "zh"
}
}
```
触发条件:`sample_mode: true`,或消息内容不包含标签且不像歌词时自动触发。
### 模式 3: 纯歌词模式
直接传入带结构标记的歌词,系统自动识别:
```json
{
"messages": [
{
"role": "user",
"content": "[Verse 1]\nWalking down the street\nFeeling the beat\n\n[Chorus]\nDance with me tonight\nUnder the moonlight"
}
],
"audio_config": {"duration": 30}
}
```
触发条件:消息内容包含 `[Verse]``[Chorus]` 等标记,或有多行短文本结构。
### 模式 4: 歌词 + Prompt 分离
通过 `lyrics` 字段直接传入歌词,messages 文本自动作为 prompt:
```json
{
"messages": [
{"role": "user", "content": "Energetic EDM with heavy bass drops"}
],
"lyrics": "[Verse 1]\nFeel the rhythm in your soul\nLet the music take control\n\n[Drop]\n(instrumental break)",
"audio_config": {
"bpm": 128,
"duration": 60
}
}
```
### 器乐模式
设置 `audio_config.instrumental: true`
```json
{
"messages": [
{"role": "user", "content": "<prompt>Epic orchestral cinematic score, dramatic and powerful</prompt>"}
],
"audio_config": {
"instrumental": true,
"duration": 30
}
}
```
---
## 音频输入
支持通过多模态 messages 传入音频文件(base64 编码),用于 cover、repaint 等任务。
### task_type 类型
| task_type | 说明 | 需要音频输入 |
|---|---|---|
| `text2music` | 文本生成音乐(默认) | 可选(作为 reference) |
| `cover` | 翻唱/风格迁移 | 需要 src_audio |
| `repaint` | 局部重绘 | 需要 src_audio |
| `lego` | 音频拼接 | 需要 src_audio |
| `extract` | 音频提取 | 需要 src_audio |
| `complete` | 音频续写 | 需要 src_audio |
### 音频路由规则
多个 `input_audio` 块按顺序路由到不同参数(类似多图片上传):
| task_type | audio[0] | audio[1] |
|---|---|---|
| `text2music` | reference_audio(风格参考) | - |
| `cover/repaint/lego/extract/complete` | src_audio(待编辑音频) | reference_audio(可选风格参考) |
### 音频输入示例
**Cover 任务(翻唱):**
```json
{
"messages": [
{
"role": "user",
"content": [
{"type": "text", "text": "<prompt>Jazz style cover with saxophone</prompt>"},
{
"type": "input_audio",
"input_audio": {"data": "<base64 原始音频>", "format": "mp3"}
}
]
}
],
"task_type": "cover",
"audio_cover_strength": 0.8,
"audio_config": {"duration": 30}
}
```
**Repaint 任务(局部重绘):**
```json
{
"messages": [
{
"role": "user",
"content": [
{"type": "text", "text": "<prompt>Replace with guitar solo</prompt>"},
{
"type": "input_audio",
"input_audio": {"data": "<base64 原始音频>", "format": "mp3"}
}
]
}
],
"task_type": "repaint",
"repainting_start": 10.0,
"repainting_end": 20.0,
"audio_config": {"duration": 30}
}
```
---
## 流式响应
设置 `"stream": true` 启用 SSE(Server-Sent Events)流式返回。
### 事件格式
每个事件以 `data: ` 开头,后跟 JSON,以双换行 `\n\n` 结尾:
```
data: {"id":"chatcmpl-xxx","object":"chat.completion.chunk","created":1706688000,"model":"acemusic/acestep-v15-turbo","choices":[{"index":0,"delta":{...},"finish_reason":null}]}
```
### 流式事件顺序
| 阶段 | delta 内容 | 说明 |
|---|---|---|
| 1. 初始化 | `{"role":"assistant","content":""}` | 建立连接 |
| 2. LM 内容 | `{"content":"\n\n## Metadata\n..."}` | LM 参与时推送 metadata 和 lyrics |
| 3. 心跳 | `{"content":"."}` | 音频生成期间每 2 秒发送,保持连接 |
| 4. 音频数据 | `{"audio":[{"type":"audio_url","audio_url":{"url":"data:..."}}]}` | 音频 base64 |
| 5. 结束 | `finish_reason: "stop"` | 生成完成 |
| 6. 终止 | `data: [DONE]` | 流结束标记 |
### 流式响应示例
```
data: {"id":"chatcmpl-abc123","object":"chat.completion.chunk","created":1706688000,"model":"acemusic/acestep-v15-turbo","choices":[{"index":0,"delta":{"role":"assistant","content":""},"finish_reason":null}]}
data: {"id":"chatcmpl-abc123","object":"chat.completion.chunk","created":1706688000,"model":"acemusic/acestep-v15-turbo","choices":[{"index":0,"delta":{"content":"\n\n## Metadata\n**Caption:** Upbeat pop\n**BPM:** 120"},"finish_reason":null}]}
data: {"id":"chatcmpl-abc123","object":"chat.completion.chunk","created":1706688000,"model":"acemusic/acestep-v15-turbo","choices":[{"index":0,"delta":{"content":"."},"finish_reason":null}]}
data: {"id":"chatcmpl-abc123","object":"chat.completion.chunk","created":1706688000,"model":"acemusic/acestep-v15-turbo","choices":[{"index":0,"delta":{"audio":[{"type":"audio_url","audio_url":{"url":"data:audio/mpeg;base64,..."}}]},"finish_reason":null}]}
data: {"id":"chatcmpl-abc123","object":"chat.completion.chunk","created":1706688000,"model":"acemusic/acestep-v15-turbo","choices":[{"index":0,"delta":{},"finish_reason":"stop"}]}
data: [DONE]
```
### 客户端处理流式响应
```python
import json
import httpx
with httpx.stream("POST", "http://127.0.0.1:8002/v1/chat/completions", json={
"messages": [{"role": "user", "content": "生成一首轻快的吉他曲"}],
"sample_mode": True,
"stream": True,
"audio_config": {"instrumental": True}
}) as response:
content_parts = []
audio_url = None
for line in response.iter_lines():
if not line or not line.startswith("data: "):
continue
if line == "data: [DONE]":
break
chunk = json.loads(line[6:])
delta = chunk["choices"][0]["delta"]
if "content" in delta and delta["content"]:
content_parts.append(delta["content"])
if "audio" in delta and delta["audio"]:
audio_url = delta["audio"][0]["audio_url"]["url"]
if chunk["choices"][0].get("finish_reason") == "stop":
print("Generation complete!")
print("Content:", "".join(content_parts))
if audio_url:
import base64
b64_data = audio_url.split(",", 1)[1]
with open("output.mp3", "wb") as f:
f.write(base64.b64decode(b64_data))
```
```javascript
const response = await fetch("http://127.0.0.1:8002/v1/chat/completions", {
method: "POST",
headers: { "Content-Type": "application/json" },
body: JSON.stringify({
messages: [{ role: "user", content: "生成一首轻快的吉他曲" }],
sample_mode: true,
stream: true,
audio_config: { instrumental: true }
})
});
const reader = response.body.getReader();
const decoder = new TextDecoder();
let audioUrl = null;
let content = "";
while (true) {
const { done, value } = await reader.read();
if (done) break;
const text = decoder.decode(value);
for (const line of text.split("\n")) {
if (!line.startsWith("data: ") || line === "data: [DONE]") continue;
const chunk = JSON.parse(line.slice(6));
const delta = chunk.choices[0].delta;
if (delta.content) content += delta.content;
if (delta.audio) audioUrl = delta.audio[0].audio_url.url;
}
}
// audioUrl 可直接用于 <audio src="...">
```
---
## 完整示例
### 示例 1: 自然语言生成(最简用法)
```bash
curl -X POST http://127.0.0.1:8002/v1/chat/completions \
-H "Content-Type: application/json" \
-d '{
"messages": [
{"role": "user", "content": "一首温柔的中文民谣,关于故乡和回忆"}
],
"sample_mode": true,
"audio_config": {"vocal_language": "zh"}
}'
```
### 示例 2: 标签模式 + 指定参数
```bash
curl -X POST http://127.0.0.1:8002/v1/chat/completions \
-H "Content-Type: application/json" \
-d '{
"messages": [
{
"role": "user",
"content": "<prompt>Energetic EDM track with heavy bass drops and synth leads</prompt><lyrics>[Verse 1]\nFeel the rhythm in your soul\nLet the music take control\n\n[Drop]\n(instrumental break)</lyrics>"
}
],
"audio_config": {
"bpm": 128,
"duration": 60,
"vocal_language": "en"
}
}'
```
### 示例 3: 纯器乐 + 关闭 LM 增强
```bash
curl -X POST http://127.0.0.1:8002/v1/chat/completions \
-H "Content-Type: application/json" \
-d '{
"messages": [
{
"role": "user",
"content": "<prompt>Peaceful piano solo, slow tempo, jazz harmony</prompt>"
}
],
"use_cot_caption": false,
"audio_config": {
"instrumental": true,
"duration": 45
}
}'
```
### 示例 4: 流式请求
```bash
curl -X POST http://127.0.0.1:8002/v1/chat/completions \
-H "Content-Type: application/json" \
-N \
-d '{
"messages": [
{"role": "user", "content": "Generate a happy birthday song"}
],
"sample_mode": true,
"stream": true
}'
```
### 示例 5: 多种子批量生成
```bash
curl -X POST http://127.0.0.1:8002/v1/chat/completions \
-H "Content-Type: application/json" \
-d '{
"messages": [
{"role": "user", "content": "<prompt>Lo-fi hip hop beat</prompt>"}
],
"batch_size": 3,
"seed": "42,123,456",
"audio_config": {
"instrumental": true,
"duration": 30
}
}'
```
---
## 错误码
| HTTP 状态码 | 说明 |
|---|---|
| 400 | 请求格式错误或缺少有效输入 |
| 401 | API Key 缺失或无效 |
| 429 | 服务繁忙,队列已满 |
| 500 | 音乐生成过程中发生内部错误 |
| 503 | 模型尚未初始化完成 |
| 504 | 生成超时 |
错误响应格式:
```json
{
"detail": "错误描述信息"
}
```
---
## 环境变量配置
以下环境变量可用于配置服务端(供运维参考):
| 变量名 | 默认值 | 说明 |
|---|---|---|
| `OPENROUTER_API_KEY` | 无 | API 认证密钥 |
| `OPENROUTER_HOST` | `127.0.0.1` | 监听地址 |
| `OPENROUTER_PORT` | `8002` | 监听端口 |
| `ACESTEP_CONFIG_PATH` | `acestep-v15-turbo` | DiT 模型配置路径 |
| `ACESTEP_DEVICE` | `auto` | 推理设备 |
| `ACESTEP_LM_MODEL_PATH` | `acestep-5Hz-lm-0.6B` | LLM 模型路径 |
| `ACESTEP_LM_BACKEND` | `vllm` | LLM 推理后端 |
| `ACESTEP_QUEUE_MAXSIZE` | `200` | 任务队列最大容量 |
| `ACESTEP_GENERATION_TIMEOUT` | `600` | 非流式请求超时(秒) |