Midjourney API

A free, open-source Midjourney API that puts Discord, YouChuan and the Midjourney official website behind one HTTP interface, with one-click face swap for images and videos.

The world's largest and most complete free open-source Midjourney API — the most powerful and full-featured, generating 1,000,000+ images per day!

⚡ Huanwang AI · Enterprise Midjourney API

Always-on FAST · 120× ultra-high concurrency · Never banned · Permanent retention · Derivatives available anytime, forever · Native MCP support

Website: https://huanwangai.com
API console: https://api.huanwangai.com

English | 中文

Documentation

Quick start

Start the service with the one-click Docker script or the published image (default port 8086).
Open the admin console, sign in, and reset the administrator token immediately.
On a private deployment, turn off demo mode, registration and guest access.
Add at least one drawing account and wait until it comes online.
Submit a minimal drawing request and confirm that the task result comes back.

⚡ Discord Official Mirror

All Discord commands supported, ready out of the box: https://trueai.org

Community

If you have questions about usage or deployment, join the community group and work through them together.

If you need drawing capacity, contact a group admin — the group is full of experienced users.

Midjourney Community Group (QQ Group: 565908696)

Backend Integration Guides

Midjourney official website: https://trueai-org.feishu.cn/docx/ZDeUdPCBho4IE0xJHxpcXbdYnKd
YouChuan: https://trueai-org.feishu.cn/docx/KrdIdOKm4otVmwxdv3AcLjsEnXf

Main Features

Discord drawing.
YouChuan drawing.
Midjourney official website drawing.
Distributed deployment.
Online upgrade.
Online restart.
Elastic deployment with load balancing and Consul as the configuration center.
Imagine command and related actions [V1/V2.../U1/U2.../R].
Base64 reference images when submitting Imagine.
Blend (image mixing), Describe (image to text) and Shorten (prompt analysis) commands.
Real-time task progress.
Chinese prompt translation, configurable through Baidu Translate or GPT.
Prompt sensitive-word pre-detection with overridable rules.
User-token WSS connection for error messages and the full command set.
Focus shift: Pan ⬅️➡⬆️⬇️.
Partial redraw: Vary (Region) 🖌.
All associated button actions.
Image zoom and custom zoom 🔍.
Retrieve the seed of a generated image.
Per-account generation speed mode: RELAX | FAST | TURBO.
Multi-account configuration with a dedicated task queue per account and selectable account strategies: BestWaitIdle | Random | Weight | Polling.
Persistent account pool with dynamic maintenance.
Fetch account /info and /settings details.
Account settings management.
Both the niji・journey Bot and the Midjourney Bot.
Safe compressed transport with zlib-stream https://discord.com/developers/docs/topics/gateway.
Embedded MJ admin console with multi-language support https://github.com/trueai-org/midjourney-proxy-webui.
Create, read, update and delete MJ accounts.
Detailed account information and account synchronization.
Per-account concurrency and queue settings.
Task queries for MJ jobs.
A complete drawing test page.
Compatible with mainstream drawing clients and API calls.
Parent task id in task payload.
🎛️ Remix mode and automatic Remix submission.
Built-in local image storage and built-in CDN acceleration.
Automatically simulates reading unread messages when too many are pending.
BOOKMARK and other commands.
Target a specific instance, filter accounts by speed, filter remix accounts and more — see the Swagger accountFilter field.
Reverse-generate task information from a job id or an image.
Account sorting, parallelism, queue count, max queue size and task interval settings.
Client path modes: default https://{BASE_URL}/mj/submit/imagine, /mj-turbo/mj for turbo mode, /mj-relax/mj for relax mode, /mj-fast/mj for fast mode, /mj for no mode.
CloudFlare manual human verification: the account is locked automatically when triggered, then verified through the GUI or an email notification.
CloudFlare automatic human verification with a configurable verification server (the auto-verifier only runs on Windows).
Working-hour windows — drawing around the clock can trigger warnings, so 8–10 hours of rest is recommended, e.g. 09:10-23:55, 13:00-08:10.
Built-in IP rate limiting, IP-range limiting, blacklists, whitelists and automatic blacklisting.
Daily drawing limit; once reached, no new jobs are accepted but variations and redraws still work.
Registration and guest access toggles.
Visual configuration.
Swagger documentation can be exposed independently.
Bot token is optional — the service works without a bot.
Improved command and progress display.
Idle-time configuration with a relaxed ("salted fish") mode per account to avoid high-frequency operation (no new drawings in that window, other commands still work, multiple time windows supported).
Vertical account categories: each account can be restricted to one kind of work, e.g. landscapes only or portraits only.
Shared channels and sub-channels: if an account is banned, earlier drawings can continue by using its channel as a sub-channel of a healthy account, with permanent invite and sub-channel links and batch editing.
SQLite (default), MySQL/MariaDB, PostgreSQL and SQL Server; about 500,000 task records are retained by default (database support sponsored by @如风).
One-click migration from mjplus or other services, including accounts and tasks.
Built-in forbidden-word management with grouped entries.
Non-official links in prompts are rewritten to official ones, so domestic or custom reference links do not trigger verification.
When fast hours run out, switch to relax mode automatically (opt-in) and restore fast mode after a top-up or renewal.
Image storage on Alibaba Cloud OSS with custom CDN, custom styles and thumbnails (OSS is recommended: separated from the origin and faster to load).
Shorten-based prompt analysis regeneration.
Image face swap — comply with local laws and never use it for illegal purposes.
Video face swap — comply with local laws and never use it for illegal purposes.
Automatic switching between relax and fast mode, plus automatic relax configuration.
No DM channel id needed — it is fetched automatically at startup.
Automatic handling of "Tos not accepted".
2FA authenticator, e.g. http://47.76.110.222:8081/code, details: http://47.76.110.222:8081/swagger/index.html.
Storage backends: Alibaba Cloud OSS, Tencent Cloud COS, local storage, S3 (thanks to @八级大狂风 for sponsoring) and Cloudflare R2 (no egress fees, see https://developers.cloudflare.com/r2/).
Independent translation configuration for MJ and NIJI.
Convert Niji to MJ: Niji・journey jobs are submitted to Midjourney with a --niji suffix, so the Niji bot is not required.
Convert --niji to the Niji Bot: prompts containing --niji are routed to the Niji・journey Bot automatically.
Automatic account login (enable 2FA on the Discord account https://github.com/trueai-org/midjourney-proxy/wiki/2FAopen and configure YesCaptchaKey on the login service https://yescaptcha.com/i/4pizLQ).
Account limits, concurrency and validity periods — thanks to @TOOM for sponsoring this feature.
Video generation.
YouChuan account import from .xlsx (up to 20 MB and 2,000 rows).
Multiple auth headers: Authorization, mj-api-secret, x-api-key and api-key.
Redis pipeline for account runtime snapshots.
HEAD probing for image URL suffix when the URL has a query string or no extension.

Online Preview

The public endpoint runs in relax mode and is free to call. The account pool is provided by sponsors — please use it responsibly.

Admin console: https://ai.trueai.org
Access: public demo mode (read-only); no credentials are published.
Public API: https://ai.trueai.org/mj
API documentation: https://ai.trueai.org/swagger (restored after the public demo site is updated; fixed locally — rely on your actual deployment)
Auto verification / auto login server: http://47.76.110.222:8081
Auto verification / auto login documentation: http://47.76.110.222:8081/swagger

The public verifier has been under heavy malicious attack, so automatic login is currently paused. Deploy your own verifier if you need it.

Preview Screenshots

Drawing

Recommended Clients

ChatAny: https://github.com/ChatAnyTeam/ChatAny
- Get your own ChatGPT + StabilityAI + Midjourney web service in one click: https://aidemo.xiazai.zip/#/mj
- Open the site → Settings → Custom API → Model (Midjourney) → API URL → https://ai.trueai.org/mj
GoAmzAI: https://github.com/Licoy/GoAmzAI
ChatGPT Web Midjourney Proxy: https://github.com/Dooy/chatgpt-web-midjourney-proxy
- Open https://vercel.ddaiai.com → Settings → MJ drawing API URL → https://ai.trueai.org
GoMaxAI: https://github.com/51yuese/GoMaxAI-ChatGPT-Midjourney-Pro
SparkAI: https://ai.sparkaigf.com

Recommended Providers

GPTNB: https://goapi.gptnb.ai
Yifeng: https://api.ephone.ai
Midjourney API: https://api.mjdjourney.cn

Installation and Usage

Note: for a private deployment, always turn off demo mode, registration and guest access so the API cannot be abused.

Note: on Windows, download the portable package and start it directly — see below.

Quick Start

Help: WIKI

Docker

Note: double-check that the mapped files and paths are correct ⚠

# Note: the official image needs more memory — 2 GB minimum on the server
# Official image (full: Discord, YouChuan and official website drawing, more dependencies)
# Installs and starts automatically
# The one-click upgrade script is recommended
# 1. First download (edit the script afterwards to customize path, port, memory, etc. — default port is 8086)
wget -O docker-upgrade.sh https://raw.githubusercontent.com/trueai-org/midjourney-proxy/main/scripts/docker-upgrade.sh && bash docker-upgrade.sh

# 2. Upgrade later by running the same script
sh docker-upgrade.sh

# 3. Docker environment variables (optional)
# Max concurrent tasks per node: -e CONCURRENT=10
# Host private IP: -e HOST_IP=10.0.0.1

# 4. Kestrel environment variables (optional)
# KESTREL_MIN_REQUEST_BODY_BYTES_PER_SECOND: minimum request body data rate in bytes per second, default 240, recommended 10 - 1000
# KESTREL_SLOW_BUFFER_SECONDS: allowed buffering time in seconds, default 5, recommended 5 - 30
# KESTREL_MAX_REQUEST_BODY_SIZE: maximum request body size in MB, default 30, recommended 10 - 100

Docker Compose documentation for all services

Database / Redis deployment and container networking

Database benchmark report

Redis: required, e.g. mjopen-redis:6379,password=***,defaultDatabase=1,prefix=mjopen:
Sqlite3: default local database, stored at data/mj_sqlite.db
MySQL8.4/MariaDB11: connection string, e.g. Data Source=mjopen-mysql;Port=3306;User ID=root;Password=***;Initial Catalog=mjopen;SslMode=none;AllowPublicKeyRetrieval=true;Min pool size=1;Default Command Timeout=120;Connection Timeout=30
SQLServer2022/2025: connection string, e.g. Data Source=mjopen-sqlserver;User Id=sa;Password=***;Initial Catalog=mjopen;Encrypt=True;TrustServerCertificate=True;Pooling=true;Min Pool Size=1
PostgreSQL16/17/18: connection string, e.g. Host=mjopen-postgres;Port=5432;Username=mj;Password=***;Database=mjopen;ArrayNullabilityMode=Always;Pooling=true;Minimum Pool Size=1

# Start with docker-compose
# Download docker-compose.yml
wget -O docker-compose.yml https://raw.githubusercontent.com/trueai-org/midjourney-proxy/main/scripts/docker-compose.yml

# Start the container and its services (edit docker-compose.yml first to customize path, port, memory, passwords, etc. — default port is 8086)
# Default: MySQL only
docker compose up -d

# Use PostgreSQL
docker compose --profile postgres up -d

# Use SQL Server (change to a strong password)
docker compose --profile sqlserver up -d

# Start several profiles at once
docker compose --profile postgres --profile sqlserver up -d

# Stop services
docker-compose down

# Follow all service logs
docker-compose logs -f

# Follow a specific service
docker-compose logs -f mjopen
docker-compose logs -f mjopen-redis
docker-compose logs -f mjopen-mysql

# Restart all services
docker-compose restart

# Restart a specific service
docker-compose restart mjopen

# Lite image (Discord and YouChuan drawing only)
docker pull registry.cn-guangzhou.aliyuncs.com/trueai-org/midjourney-proxy:lite
docker pull registry.cn-guangzhou.aliyuncs.com/trueai-org/midjourney-proxy:lite-{version}

# Manual install and start
# Alibaba Cloud image (recommended in mainland China)
docker pull registry.cn-guangzhou.aliyuncs.com/trueai-org/midjourney-proxy

# 1. Business configuration is written to `data/mj.json` by the admin console; no appsettings template download is needed.

# 2. Stop and remove the old container
docker stop mjopen && docker rm mjopen

# 3. Start the container
docker run --name mjopen -d --restart=always \
 -p 8086:8080 --user root \
 -v /root/mjopen/logs:/app/logs:rw \
 -v /root/mjopen/data:/app/data:rw \
 -v /root/mjopen/attachments:/app/wwwroot/attachments:rw \
 -v /root/mjopen/ephemeral-attachments:/app/wwwroot/ephemeral-attachments:rw \
 -e TZ=Asia/Shanghai \
 -v /etc/localtime:/etc/localtime:ro \
 -v /etc/timezone:/etc/timezone:ro \
 registry.cn-guangzhou.aliyuncs.com/trueai-org/midjourney-proxy

# GitHub mirror
docker pull ghcr.io/trueai-org/midjourney-proxy
docker run --name mjopen -d --restart=always \
 -p 8086:8080 --user root \
 -v /root/mjopen/logs:/app/logs:rw \
 -v /root/mjopen/data:/app/data:rw \
 -v /root/mjopen/attachments:/app/wwwroot/attachments:rw \
 -v /root/mjopen/ephemeral-attachments:/app/wwwroot/ephemeral-attachments:rw \
 -e TZ=Asia/Shanghai \
 -v /etc/localtime:/etc/localtime:ro \
 -v /etc/timezone:/etc/timezone:ro \
 ghcr.io/trueai-org/midjourney-proxy

# Docker Hub mirror
docker pull trueaiorg/midjourney-proxy
docker run --name mjopen -d --restart=always \
 -p 8086:8080 --user root \
 -v /root/mjopen/logs:/app/logs:rw \
 -v /root/mjopen/data:/app/data:rw \
 -v /root/mjopen/attachments:/app/wwwroot/attachments:rw \
 -v /root/mjopen/ephemeral-attachments:/app/wwwroot/ephemeral-attachments:rw \
 -e TZ=Asia/Shanghai \
 -v /etc/localtime:/etc/localtime:ro \
 -v /etc/timezone:/etc/timezone:ro \
 trueaiorg/midjourney-proxy

Windows (ARM builds are no longer supported)

a. Download the latest portable build for Windows from https://github.com/trueai-org/midjourney-proxy/releases, e.g. midjourney-proxy-win-x64.zip
b. Extract it and run Midjourney.API.exe
c. Open http://localhost:8080
d. (Optional) Deploy to IIS: add the folder as a website, set the application pool to "No Managed Code" and start the site.
e. (Optional) Use the built-in Task Scheduler to create a basic task that runs the .exe, and choose "Do not start a new instance" so only one instance runs.

Linux (ARM builds are no longer supported)

a. Download the latest portable build for Linux from https://github.com/trueai-org/midjourney-proxy/releases, e.g. midjourney-proxy-linux-x64.zip
b. Extract it in the current directory: tar -xzf midjourney-proxy-linux-x64-<VERSION>.tar.gz
c. The entry point is run_app.sh
c. Start method 1: sh run_app.sh
d. Start method 2: chmod +x run_app.sh && ./run_app.sh

macOS (ARM builds are no longer supported)

a. Download the latest portable build for macOS from https://github.com/trueai-org/midjourney-proxy/releases, e.g. midjourney-proxy-osx-x64.zip
b. Extract it in the current directory: tar -xzf midjourney-proxy-osx-x64-<VERSION>.tar.gz
c. The entry point is run_app_osx.sh
c. Start method 1: sh run_app_osx.sh
d. Start method 2: chmod +x run_app_osx.sh && ./run_app_osx.sh

Linux one-click install script (❤ thanks to @dbccccccc)

# Method 1
wget -N --no-check-certificate https://raw.githubusercontent.com/trueai-org/midjourney-proxy/main/scripts/linux_install.sh && chmod +x linux_install.sh && bash linux_install.sh

# Method 2
curl -o linux_install.sh https://raw.githubusercontent.com/trueai-org/midjourney-proxy/main/scripts/linux_install.sh && chmod +x linux_install.sh && bash linux_install.sh

Paths

/app/data — data directory for accounts, tasks and other state
- /app/data/mj.json — configuration file
/app/logs — log directory
/app/wwwroot — static files
- /app/wwwroot/attachments — generated drawings
- /app/wwwroot/ephemeral-attachments — images produced by describe

Roles

Regular user: drawing API only, cannot sign in to the console.
Admin: can sign in to the console, inspect tasks and change settings.

Default User

Inject a strong administrator token through AdminToken before first startup; rotate it immediately after the first login.

Alibaba Cloud OSS Configuration

{
  "bucketName": "mjopen", // OSS bucket name
  "accessKeyId": "LTAI***", // OSS access key id
  "accessKeySecret": "<YOUR_OSS_SECRET>", // OSS access key secret
  "endpoint": "oss-cn-hongkong-internal.aliyuncs.com", // OSS endpoint
  "customCdn": "custom CDN domain",
  "imageStyle": null,
  "thumbnailImageStyle": null,
  "videoSnapshotStyle": null,
  "expiredMinutes": 0
}

Face Swap Configuration

{
  "token": "****",
  "enableFaceSwap": true,
  "faceSwapVersion": "d1d6ea8c8be89d664a07a457526f7128109dee7030fdac424788d762c71ed111",
  "faceSwapCoreSize": 3,
  "faceSwapQueueSize": 10,
  "faceSwapTimeoutMinutes": 10,
  "enableVideoFaceSwap": true,
  "videoFaceSwapVersion": "104b4a39315349db50880757bc8c1c996c5309e3aa11286b0a3c84dab81fd440",
  "videoFaceSwapCoreSize": 3,
  "videoFaceSwapQueueSize": 10,
  "videoFaceSwapTimeoutMinutes": 30,
  "maxFileSize": 10485760,
  "webhook": null,
  "webhookEventsFilter": []
}

Banned Prompt Rate Limiting

Configures how long (in minutes) a user is blocked after Banned prompt detected is triggered n times in one day. Whitelisted users are excluded.

{
  "enable": true,
  "rules": {
    "1": 60,
    "2": 120,
    "3": 600
  }
}

Bot Token (optional)

The bot token is used to connect to Discord over WSS and gives you error messages and the full feature set.

Bot token configuration guide

Development

Requirements: .NET 8 SDK; the solution targets net8.0.

# Restore and build
dotnet restore Midjourney.sln
dotnet build Midjourney.sln

# Run the API (Development profile)
dotnet run --project src/Midjourney.API/Midjourney.API.csproj

# Run the tests
dotnet test src/Midjourney.Tests/Midjourney.Tests.csproj

Logs are written to the logs directory next to the executable; inside the container they go to /app/logs.
With the Development profile the console and Swagger UI are served on http://localhost:5229; in Docker the service listens on 8080, mapped to 8086 by the commands above.

CloudFlare CAPTCHA Verification

Free automatic CAPTCHA handling. Example CloudFlare auto-verification configuration:

CaptchaServer: address of the verification server.
CaptchaNotifyHook: callback fired when verification completes; defaults to your domain.

"CaptchaServer": "http://47.76.110.222:8081",
"CaptchaNotifyHook": "https://ai.trueai.org"

CloudFlare Verifier / Auto Login

Windows only (TLS 1.3 supported; requires Windows 11 or Windows Server 2022). The verifier drives Chrome, so it has to run on Windows — a Linux deployment would need too many extra libraries.

Note: 2captcha is currently unstable. Self-hosting requires a 2captcha.com API key, otherwise it will not work. Pricing: 1000 verifications / 9 CNY, site: https://2captcha.cn/p/cloudflare-turnstile.

Tip: the first startup downloads Chrome, which can take a while.

Sample verifier appsettings.json (verifier service configuration, not the main application configuration)

{
  "Captcha": {
    "IsLoginService": true, // enable the login service
    "Concurrent": 1, // concurrency
    "Headless": true, // run Chrome headless
    "TwoCaptchaKey": "", // 2captcha.com API key
    "YesCaptchaKey": "" // yescaptcha.com API key
  },
  "urls": "http://*:8081" // default port
}

Preventing Warning on Frequent Drawing

Keep the task interval between 30 and 180 seconds, with at least 3.6 seconds before execution.
At most 200 images per day.
Recommended working hours: 09:10–22:50.
With multiple accounts, enable vertical categories so each account only produces one kind of work.

Consul Configuration

Technical layout

Caddy -> [Yarp | Traefik] -> Midjourney API -> Consul <-> [Yarp | Traefik]

Multiple nodes share the same Redis; Consul serves as the configuration center and for service discovery; the gateway can be YARP or Traefik. Node IPs, ports and tokens depend on the deployment environment.


# Stop and remove the old container
docker stop consul && docker rm consul

# Host network (sample node address: 192.168.3.241)
docker run -d --name consul --restart always --network host hashicorp/consul agent -server -ui -node=server-1 -bootstrap-expect=1 -client=0.0.0.0 -bind=192.168.3.241

# Port mapping
docker run -d --name consul --restart always -p 8500:8500 -p 8600:8600/udp hashicorp/consul agent -server -ui -node=server-1 -bootstrap-expect=1 -client=0.0.0.0 -bind=0.0.0.0

S3 - MinIO

https://min.io/docs/minio/linux/reference/minio-mc-admin/mc-admin-user.html

Example

# Start the container
docker run -d -p 9000:9000 -p 9001:9001 \
  --name minio-test \
  -e "MINIO_ROOT_USER=<YOUR_MINIO_USER>" \
  -e "MINIO_ROOT_PASSWORD=<YOUR_MINIO_PASSWORD>" \
  quay.io/minio/minio:latest server /data --console-address ":9001"

# Set an mc alias that points to your MinIO service first.
mc alias set myminio http://192.168.3.241:9000 "<YOUR_MINIO_USER>" "<YOUR_MINIO_PASSWORD>"

# Verify
mc ls myminio

# Create the test bucket if it does not exist
mc mb myminio/test

# Allow anonymous read-only access
mc anonymous set download myminio/test

Image Acceleration Notes (Reference Only)

How do you choose the right acceleration for different client needs?

Test tool: https://boce.aliyun.com/detect/http

Users in mainland China: a domestic CDN is the fastest, most stable and cheapest option.
Users abroad: a global CDN is faster, more stable and cheaper.

Midjourney CDN, behind the CloudFlare firewall, ~3s on average, may time out
https://cdn.midjourney.com/4872b862-7aeb-4346-a87d-16e077a2935d/0_2.png

Origin acceleration via OSS bucket (0.5 CNY/GB), 2s ~ 3s on average, may time out
https://mjopen.oss-cn-hongkong.aliyuncs.com/attachments/4872b862-7aeb-4346-a87d-16e077a2935d/0_2.png

Transfer acceleration (global CDN, 1 TB/126 CNY, reachable at home and abroad, slightly slower in mainland China), 1s ~ 3s on average, usually responds
https://mjcdn.googlec.cc/attachments/4872b862-7aeb-4346-a87d-16e077a2935d/0_2.png

Transfer acceleration (domestic CDN, 1 TB/126 CNY, fastest in mainland China), 30ms ~ 200ms on average
https://mjcn-midjourney.googlec.cc/attachments/4872b862-7aeb-4346-a87d-16e077a2935d/0_2.png

Transfer acceleration (0.5 ~ 1.25 CNY/GB, dynamic acceleration, not CDN nodes), 150ms ~ 500ms on average
https://mjopen.oss-accelerate.aliyuncs.com/attachments/4872b862-7aeb-4346-a87d-16e077a2935d/0_2.png

Tencent Cloud EdgeOne, free plan, unlimited traffic, test only, 300ms ~ 1s on average
http://img.aitop3000.com/attachments/4872b862-7aeb-4346-a87d-16e077a2935d/0_2.png

Other options: a proxy, self-hosted storage, local storage, etc.

Support and Sponsorship

If this project helps you, please give it a Star ⭐.
You can also donate an idle drawing account (one relax queue) to support the project 😀.
For ideas and feedback, open a PR, an Issue or a Discussion, or join the QQ group.

Security Policy

Some open-source authors have faced legal consequences, so this project must not be used for illegal purposes.

Follow national laws. Any illegal use is the sole responsibility of the user.
This project is released under the GPL. Personal and commercial use are allowed, but permission from the author is required and copyright notices must be kept.
Follow local laws and regulations; do not use it for illegal purposes.
Do not use it for illegal activities.

midjourney-proxy