hailuo free api
0.0.16
Supports high-speed streaming output, multiple rounds of dialogue, speech synthesis, speech recognition, Internet search, long document interpretation, image analysis, zero-configuration deployment, multi-channel token support, and automatic cleaning of session traces.
Fully compatible with ChatGPT interface.
There are also the following eight free-api welcome to pay attention to:
Moonshot AI (Kimi.ai) interface to API kimi-free-api
ZhipuAI (Zhipu Qingyan) interface to API glm-free-api
StepChat interface to API step-free-api
Alibaba Tongyi (Qwen) interface to API qwen-free-api
Metaso AI (Metaso) interface to API metaso-free-api
iFlytek Spark interface to API spark-free-api
DeepSeek interface to API deepseek-free-api
Listening Intelligence (Emohaa) interface to API emohaa-free-api
The reverse API is unstable. It is recommended to go to the official MiniMax https://www.minimaxi.com/platform to pay to use the API to avoid the risk of being banned.
This organization and individuals do not accept any financial donations or transactions. This project is purely for research, exchange and learning!
It is for personal use only, and it is prohibited to provide external services or commercial use to avoid putting pressure on the official service, otherwise it is at your own risk!
It is for personal use only, and it is prohibited to provide external services or commercial use to avoid putting pressure on the official service, otherwise it is at your own risk!
It is for personal use only, and it is prohibited to provide external services or commercial use to avoid putting pressure on the official service, otherwise it is at your own risk!
This link is only a temporary test function, with only one concurrency. If you encounter an exception, please try again later. It is recommended to deploy it yourself.
https://udify.app/chat/uqBly3aW1LTwzzb3
Speech synthesis is creating speech
Speech recognition is all about creating transcripts
Get token from Conch AI
Enter Conch AI to initiate a conversation, then open the developer tools with F12 and find the value of _token from Application > LocalStorage. This will be used as the Bearer Token value of Authorization: Authorization: Bearer TOKEN
At present, it seems that the same account can only have one output at the same time. You can provide the _tokens of multiple accounts and use them , the splicing provides:
Authorization: Bearer TOKEN1,TOKEN2,TOKEN3
One of these will be selected each time the service is requested.
Please prepare a server with a public IP and open port 8000.
Pull the image and start the service
docker run -it -d --init --name hailuo-free-api -p 8000:8000 -e TZ=Asia/Shanghai vinlic/hailuo-free-api:latestView real-time service logs
docker logs -f hailuo-free-apiRestart service
docker restart hailuo-free-apiStop service
docker stop hailuo-free-api version : ' 3 '
services :
hailuo-free-api :
container_name : hailuo-free-api
image : vinlic/hailuo-free-api:latest
restart : always
ports :
- " 8000:8000 "
environment :
- TZ=Asia/ShanghaiNote: Some deployment areas may not be able to connect to Hailuo. If the container log shows a request timeout or cannot be connected, please switch to other areas for deployment! Note: The container instance of the free account will automatically stop running after a period of inactivity, which will cause a delay of 50 seconds or more on the next request. It is recommended to check Render container keep alive
Fork this project to your github account.
Visit Render and log in to your github account.
Build your Web Service (New+ -> Build and deploy from a Git repository -> Connect your forked project -> Select the deployment area -> Select the instance type as Free -> Create Web Service).
After waiting for the construction to complete, copy the assigned domain name and splice the URL for access.
Note: The request response timeout for Vercel free accounts is 10 seconds, but the interface response usually takes longer, and you may encounter a 504 timeout error returned by Vercel!
Please make sure you have installed the Node.js environment first.
npm i -g vercel --registry http://registry.npmmirror.com
vercel login
git clone https://github.com/LLM-Red-Team/hailuo-free-api
cd hailuo-free-api
vercel --prodPlease prepare a server with a public IP and open port 8000.
Please install the Node.js environment and configure the environment variables first, and confirm that the node command is available.
Install dependencies
npm iInstall PM2 for process guarding
npm i -g pm2Compile and build. When you see the dist directory, the build is complete.
npm run buildStart service
pm2 start dist/index.js --name " hailuo-free-api "View real-time service logs
pm2 logs hailuo-free-apiRestart service
pm2 reload hailuo-free-apiStop service
pm2 stop hailuo-free-apiIt is faster and easier to use the following secondary development client to access the free-api series projects, and supports document/image upload!
LobeChat developed by Clivia https://github.com/Yanyutin753/lobe-chat
ChatGPT Web https://github.com/SuYxh/chatgpt-web-sea developed by Guangguang@
Currently, the /v1/chat/completions interface that is compatible with openai is supported. You can use the client access interface that is compatible with openai or other clients, or use online services such as dify to access it.
Conversation completion interface, compatible with openai's chat-completions-api.
POST /v1/chat/completions
header needs to set the Authorization header:
Authorization: Bearer [_token]
Request data:
{
// model模型名称可以乱填
"model" : " hailuo " ,
"messages" : [
{
"role" : " user " ,
"content" : "你是谁？ "
}
],
// 如果使用SSE流请设置为true，默认false
"stream" : false
}Response data:
{
"id" : " 242830597915504644 " ,
"model" : " hailuo " ,
"object" : " chat.completion " ,
"choices" : [
{
"index" : 0 ,
"message" : {
"role" : " assistant " ,
"content" : "我是海螺AI，由上海稀宇科技有限公司（MiniMax）自主研发的AI助理。我可以帮助你回答各种问题，提供信息查询、生活建议、学习辅导等服务。如果你有任何问题，随时可以向我提问。 "
},
"finish_reason" : " stop "
}
],
"usage" : {
"prompt_tokens" : 1 ,
"completion_tokens" : 1 ,
"total_tokens" : 2
},
"created" : 1714751470
}Create a voice interface, compatible with openai's audio-create-speech-api, and only supports mp3 format output.
male-botong 思远 [兼容 tts-1 alloy]
Podcast_girl 心悦 [兼容 tts-1 echo]
boyan_new_hailuo 子轩 [兼容 tts-1 fable]
female-shaonv 灵儿 [兼容 tts-1 onyx]
YaeMiko_hailuo 语嫣 [兼容 tts-1 nova]
xiaoyi_mix_hailuo 少泽 [兼容 tts-1 shimmer]
xiaomo_sft 芷溪 [兼容 tts-1-hd alloy]
cove_test2_hailuo 浩翔（英文）
scarlett_hailuo 雅涵（英文）
Leishen2_hailuo 模仿雷电将军 [兼容 tts-1-hd echo]
Zhongli_hailuo 模仿钟离 [兼容 tts-1-hd fable]
Paimeng_hailuo 模仿派蒙 [兼容 tts-1-hd onyx]
keli_hailuo 模仿可莉 [兼容 tts-1-hd nova]
Hutao_hailuo 模仿胡桃 [兼容 tts-1-hd shimmer]
Xionger_hailuo 模仿熊二
Haimian_hailuo 模仿海绵宝宝
Robot_hunter_hailuo 模仿变形金刚
Linzhiling_hailuo 小玲玲
huafei_hailuo 拽妃
lingfeng_hailuo 东北er
male_dongbei_hailuo 老铁
Beijing_hailuo 北京er
JayChou_hailuo JayJay
Daniel_hailuo 潇然
Bingjiao_zongcai_hailuo 沉韵
female-yaoyao-hd 瑶瑶
murong_sft 晨曦
shangshen_sft 沐珊
kongchen_sft 祁辰
shenteng2_hailuo 夏洛特
Guodegang_hailuo 郭嘚嘚
yueyue_hailuo 小月月
Find the robot_custom_config request response data.formInfo.userVoiceList's own clone voice ID from F12 Network, in the format of puv_****************** .
Please configure the REPLACE_AUDIO_MODEL environment variable in docker startup or the system to map custom speakers to openai's speaker model. Use commas , separate each speaker.
The order of mapping is: alloy, echo, fable, onyx, nova, shimmer. Currently, 6 mappings can be defined.
Mapping example:
Podcast_girl -> alloy
yueyue_hailuo -> echo
keli_hailuo -> fable
The environment configuration for the above mapping relationship is as follows:
REPLACE_AUDIO_MODEL="Podcast_girl,yueyue_hailuo,keli_hailuo";
POST /v1/audio/speech
header needs to set the Authorization header:
Authorization: Bearer [_token]
Request data:
{
// model模型名称可以乱填
"model" : " hailuo " ,
// 语音内容，尽量不要包含指令（否则可能导致模型回答你的问题）
"input" : "你在做什么？ " ,
// 发音人ID，可以使用官方或者自己克隆的音色
"voice" : " Podcast_girl "
}Response data:
audio/mpeg binary data stream (mp3 file)
Create a transcription interface, compatible with openai's audio-create-transcription-api.
POST /v1/audio/transcriptions
header needs to set the Authorization header:
Authorization: Bearer [_token]
Request data (multipart/form-data):
file 要转录的音频文件对象(不是文件名)，格式为:wav、mp3、mpeg、mpga、m4a、ogg、flac。
model 模型名称，可以乱填
response_format 仅支持json或text
Response data:
{
"text": "嗯，多年前呢我是个穷小子，我有一个喜欢的女孩，他有一双会说话的眼睛，他偏爱雏菊般的淡黄色。我我每天都骑自行车送他上下学专挑那个坑坑洼洼的路走。然后编的时候，我就能感觉到他用双手在后边用力的拽我的衣服，我好开心哪。然后回家以后，我才发现我唯一的一件衬衣变成了燕尾服。我每天中午会把我妈妈给我带的荷包蛋，我给她吃，他只吃蛋白，他把蛋蛋黄留给我，我我真的我好感动啊。呃，到后来我我知道呃，吃蛋黄反腹唇口，呃，我们同窗三年，我给他写了一百多封信，跟你联系我我我我我我每一封信我都换一个笔记，我怕他认出来，我那我很不好意思。所以说长此以往的练习，我在书法大赛获得了一等奖。直到有一天，他准备坐火车去省城上学的时候，我也没有把我自己的话跟他说出来。哎，我去那天也是像今天一样下着雨，他也带着那条漂亮的黄丝巾，我递给他一篮子鸡蛋，他没接，他反问我说有多少个鸡蛋，我说有一百个，他说他一天吃一个，一百天就吃完了再想吃，还有吗？这个其实我知道他懂我那句话，可是我当时就像被雷击中了一样无果果主人我什么话我都说不出来。当我缓过神儿来的时候，他已经上车了，车已经开了，车已经开出一段距离了。那个时候我只要大声的喊，他一定能够听到我张嘴了，没声儿。从那一刻起，我就生了病，预言是癌症，没当下雨或者是重要场合，我者无法说话。其实这么多年我一直想找一个机会，我希望他能够重新出现在我眼前。今天这个机会来了，此时此刻他就坐在离我三米半远的地方。我要把我二十年前要跟你说的话，大声的告诉你，哇，李丽，你你只要愿意跟我在一起鸡蛋，我给你带一辈子，有的是。"
}Provide an accessible file URL or BASE64_URL to parse.
POST /v1/chat/completions
header needs to set the Authorization header:
Authorization: Bearer [_token]
Request data:
{
// model模型名称可以乱填
"model" : " hailuo " ,
"messages" : [
{
"role" : " user " ,
"content" : [
{
"type" : " file " ,
"file_url" : {
"url" : " https://mj101-1317487292.cos.ap-shanghai.myqcloud.com/ai/test.pdf "
}
},
{
"type" : " text " ,
"text" : "文档里说了什么？ "
}
]
}
],
// 如果使用SSE流请设置为true，默认false
"stream" : false
}Response data:
{
"id" : " 242835041910616068 " ,
"model" : " hailuo " ,
"object" : " chat.completion " ,
"choices" : [
{
"index" : 0 ,
"message" : {
"role" : " assistant " ,
"content": "文档中包含了一系列的古代魔法仪式和咒语，这些内容似乎源自古希腊罗马时期的魔法文献，如《希腊魔法纸莎草纸》（PGM，全称为Papyri Graecae Magicae）。以下是每个文档内容的简要概述：nn1. 文档1中描述了一个仪式，其中包括将面包分成七小块，然后在那些被暴力杀害的地方留下这些面包块，并念诵一段咒语。这个仪式的目的是为了吸引一个特定女性的注意，让她对施法者产生强烈的感情和欲望。nn2. 文档2中包含了一个咒语，要求一个名为Didymos的施法者召唤一个名为Tereous的女性，通过念诵一系列的魔法名字和咒语，使她感到痛苦和渴望，直到她来到施法者身边。nn3. 文档3中提供了一个简单的仪式，施法者需要保持三天的纯洁，并献上乳香作为祭品，念诵一个特定的名字（NEPHERIRI），以此来吸引一个美丽的女性。nn4. 文档4中描述了一个使用没药的仪式，施法者在献上没药的同时念诵一段咒语，目的是让一个特定的女性对施法者产生强烈的爱慕之情，即使她正在做其他事情，也会被这种强烈的感情所占据。nn这些文档内容反映了古代人们对于魔法和咒语的信仰，以及他们试图通过这些仪式来影响他人情感和行为的愿望。需要注意的是，这些内容仅供学术研究和了解历史之用，现代社会中不应使用这些仪式或咒语来干预他人的自由意志。"
},
"finish_reason" : " stop "
}
],
"usage" : {
"prompt_tokens" : 1 ,
"completion_tokens" : 1 ,
"total_tokens" : 2
},
"created" : 1714752530
}Provide an accessible image URL or BASE64_URL to parse.
This format is compatible with the gpt-4-vision-preview API format, and you can also use this format to transmit documents for parsing.
POST /v1/chat/completions
header needs to set the Authorization header:
Authorization: Bearer [_token]
Request data:
{
"model" : " hailuo " ,
"messages" : [
{
"role" : " user " ,
"content" : [
{
"type" : " image_url " ,
"image_url" : {
"url" : " https://ecmb.bdimg.com/tam-ogel/-1384175475_-1668929744_259_194.jpg "
}
},
{
"type" : " text " ,
"text" : "图里是什么？ "
}
]
}
],
// 如果使用SSE流请设置为true，默认false
"stream" : false
}Response data:
{
"id" : " 242835404705341445 " ,
"model" : " hailuo " ,
"object" : " chat.completion " ,
"choices" : [
{
"index" : 0 ,
"message" : {
"role" : " assistant " ,
"content" : "图里是“海螺AI”的标识。 "
},
"finish_reason" : " stop "
}
],
"usage" : {
"prompt_tokens" : 1 ,
"completion_tokens" : 1 ,
"total_tokens" : 2
},
"created" : 1714752616
}Check whether _token is alive. If live is not true, otherwise it is false. Please do not call this interface frequently (less than 10 minutes).
POST /token/check
Request data:
{
"token" : " eyJhbGciOiJIUzUxMiIsInR5cCI6IkpXVCJ9... "
}Response data:
{
"live" : true
}If you are using Nginx reverse proxy hailuo-free-api, please add the following configuration items to optimize the output effect of the stream and optimize the experience.
# 关闭代理缓冲。当设置为off时，Nginx会立即将客户端请求发送到后端服务器，并立即将从后端服务器接收到的响应发送回客户端。
proxy_buffering off ;
# 启用分块传输编码。分块传输编码允许服务器为动态生成的内容分块发送数据，而不需要预先知道内容的大小。
chunked_transfer_encoding on ;
# 开启TCP_NOPUSH，这告诉Nginx在数据包发送到客户端之前，尽可能地发送数据。这通常在sendfile使用时配合使用，可以提高网络效率。
tcp_nopush on ;
# 开启TCP_NODELAY，这告诉Nginx不延迟发送数据，立即发送小数据包。在某些情况下，这可以减少网络的延迟。
tcp_nodelay on ;
# 设置保持连接的超时时间，这里设置为120秒。如果在这段时间内，客户端和服务器之间没有进一步的通信，连接将被关闭。
keepalive_timeout 120 ;Since the inference side is not in hailuo-free-api, the token cannot be counted and will be returned as a fixed number.
