aigc-cli：终端原生的多厂商 AIGC 工具包，内置 MCP Server

Terminal-native AIGC toolkit: multi-provider image/video/audio generation, Midjourney, chat, AIGC forensics, knowledge base, MCP Server.

中文版文档

Generate, detect, and manage AI content from the terminal. Supports OpenAI, OpenRouter, Anthropic, Ollama and any compatible relay — with local offline models for AIGC detection, OCR, background removal, and knowledge base.

Why aigc-cli?

Quick Start

Installation

Download the binary for your platform from the Releases page.

Using OpenAI

export OPENAI_API_KEY="sk-xxx"

aigc-cli image --prompt "a cat under the stars"
aigc-cli chat --message "Hello, who are you?"

Using OpenRouter (change env vars, commands stay the same)

export OPENAI_API_KEY="sk-or-xxx"
export OPENAI_BASE_URL="https://openrouter.ai/api/v1"

aigc-cli image --model "openai/gpt-image-2" --prompt "a cat"
aigc-cli video --model "google/veo-3.1" --prompt "a dog running"     # auto-routes to dedicated video API
aigc-cli models --type image                                           # authentication-free model discovery

Using any OpenAI-Compatible Relay

export OPENAI_API_KEY="sk-xxx"
export OPENAI_BASE_URL="https://your-relay.com/v1"

aigc-cli chat --message "Hello"

MCP Integration (Recommended)

Add to your MCP config in Claude Desktop / Cursor / Windsurf:

{
  "mcpServers": {
    "aigc-cli": {
      "command": "aigc-cli",
      "args": ["mcp"]
    }
  }
}

AI agents can generate images/videos/music, run Midjourney, OCR, search the idea library or knowledge base, query model pricing, and detect AIGC directly in conversation — via 30 tools, 4 workflow prompts, and 4 read-only resources (all tools carry MCP annotations). The per-call provider argument is restricted to a whitelist of names from config.providers; no tool accepts base_url or api_key. See docs/en/guide-mcp.md.

Feature Overview

Provider Auto-Adaptation

The same image / video / audio / music / models command automatically uses the correct API path based on the provider:

Bailian (阿里云百炼) currently has a dedicated route for music only — Fun-Music uses the DashScope native services endpoint, which differs from the compatible-mode path used for chat. Image / video / audio fall through to the generic OpenAI-compatible path.

Local models/services don't need an API Key. aigc-cli auto-exempts API Key checks and skips the Authorization header. See docs/en/installation.md#local-generation.

Detection logic: auto-identifies by base_url, or manually specify with --mode sync / --mode async. Each command can use a different provider via the providers config, see docs/en/examples/config.yaml.

Commands

aigc-cli
├── image / img   Image generation (sync/async/OpenRouter dedicated API/Grok Edit)  →  docs/en/guide-image.md
├── video / vid   Video generation (Agnes / OpenRouter / OpenLux + VEO3 Remix)          →  docs/en/guide-video.md
├── depth         Image/video → grayscale depth map (offline ONNX, V2 models)   →  docs/en/guide-depth.md
├── upscale / sr  Image super-resolution 2x/4x (offline ONNX: Real-ESRGAN/Real-CUGAN/Swin2SR)  →  docs/en/guide-upscale.md
├── audio / voice Audio: TTS and STT                                                →  docs/en/guide-audio.md
│   ├── tts / speak  Text-to-speech (cloud API or local sherpa-onnx offline)
│   ├── asr / stt    Speech-to-text (cloud API or local sherpa-onnx offline)
│   └── init         Download local models (kokoro, sense-voice, etc.)
├── music      AI music generation (APIMart suno/flowmusic async, OpenRouter Lyria sync, Bailian Fun-Music)  →  docs/en/guide-music.md
│   ├── generate / gen  Submit a music generation task
│   └── query           Query music task status and result
├── ocr            Offline text recognition (DBNet + CRNN, ONNX local inference)     →  docs/en/guide-ocr.md
│   ├── init        Download OCR models
│   └── scan        Recognize text in images
│                   --engine auto|pdf|ocr  PDF processing mode (default: auto)
│                   --preview              Preview recognized text
├── background / bg AI background removal (RMBG 2.0 semantic segmentation, offline ONNX)  →  docs/en/guide-background.md
├── midjourney / mj                                                                  →  docs/en/guide-midjourney.md
│   └── mj     Alias for midjourney
├── chat      AI chat / Interactive REPL / Agent Loop (tool calling)                  →  docs/en/guide-chat.md
├── ideas / idea  Prompt idea search (local dataset + online libraries)                →  docs/en/guide-ideas.md
├── knowledgebase / kb  Local knowledge base (FTS5 + semantic search, pluggable embeddings) →  docs/en/guide-knowledgebase.md
├── models / model                                                                    →  docs/en/guide-commands.md
│   └── --price    View model pricing
├── task       Query async task status (APIMart compatible)
├── balance    Query balance (APIMart compatible)
├── config     Read/edit config.yaml (get/set/list; secrets masked & auto-encrypted)  →  docs/en/guide-commands.md
├── preview / pr View images / --detail metadata / --describe caption                 →  docs/en/guide-preview.md
├── detect     Detect AIGC, metadata and tampering (multi-signal fusion + emoji)     →  docs/en/guide-detect.md
├── decision / decide  Typed decision models (System One / Jev): state + question bank → answers   →  docs/en/guide-decision.md
├── completion Generate shell completion scripts (bash/zsh/fish/powershell)
├── mcp        Start MCP Server (AI agent integration)                                →  docs/en/guide-mcp.md
│
│   # Global flags
│   --dry-run      Print request params and equivalent curl, no API call
│   --print-config Print effective config with source annotations (secrets masked)
│   -v/--verbose   Show detailed output: full JSON, token usage, timing, cost
│   --json         Pass request as JSON (file, string, or stdin)
│   --preview      Open system preview after generation
│   --save-prompt  Save prompt as .md file
│   --http-proxy   Specify HTTP proxy
│   --zdr          Request zero data retention where supported (OpenRouter chat/music)

Enable Tab Completion

After installation, add one of the following lines to your shell rc file:

# Bash
echo 'source <(aigc-cli completion bash)' >> ~/.bashrc

# Zsh
echo 'source <(aigc-cli completion zsh)' >> ~/.zshrc

# Fish
aigc-cli completion fish > ~/.config/fish/completions/aigc-cli.fish

After reloading your shell, typing aigc-cli im+Tab will auto-complete to aigc-cli image.

Midjourney Subcommands

aigc-cli midjourney (or mj)
├── imagine       Text-to-image / Image-to-image (default entry)
├── blend         Multi-image blend (2-4 images)
├── describe      Image-to-text (reverse prompt)
├── edits         Image editing (rewrite entire image)
├── upscale       Upscale single image (U1-U4)
├── variation     Mild variation (V1-V4)
├── high-variation  Strong variation
├── low-variation   Weak variation
├── reroll        Regenerate grid
├── zoom          Zoom out / out-paint
├── pan           Pan (left/right/up/down)
├── inpaint       Local repaint entry (→ modal)
├── modal         Submit mask + prompt to complete repaint
├── video         Image-to-video
├── remix-strong  Strong remix (v8/v8.1)
├── remix-subtle  Weak remix (v8/v8.1)
└── query         Query task status

Documentation

Priority Rules

CLI flags > JSON input > YAML config > Code defaults

For --json this precedence holds per key: when flags accompany --json, each flag you explicitly set overrides the matching key in the JSON body, while every key you did not touch is kept exactly as written. With --json and no flags, the body is sent verbatim.

--json also accepts JSONC: // and /* */ comments, trailing commas, and a leading UTF-8 BOM are stripped before parsing (string-literal aware, so the // in a URL is kept). Strict JSON with no comments passes through unchanged.

# JSON holds the stable params (model, loras, seed, steps, size); the prompt varies per run
aigc-cli image --provider modelscope --json downloads/prompt.json --prompt "a low-angle legwear ad"

Proxy priority: --http-proxy flag > HTTP_PROXY / HTTPS_PROXY standard env vars

Contributing

Contributions welcome! See CONTRIBUTING.md.

Legal Notice

Users must ensure their usage complies with applicable laws and regulations. The software authors assume no legal liability for any user actions.

License

MIT