undefined

**Restart Claude Code** after installation. Then run the setup wizard:

```bash
/meigen:setup

undefined

undefined

undefined

undefined

{
  "mcpServers": {
    "meigen": {
      "command": "npx",
      "args": ["-y", "meigen@1.3.1"],
      "env": {
        "MEIGEN_API_TOKEN": "meigen_sk_..."
      }
    }
  }
}

undefined

undefined

undefined

undefined

{
  "provider": "meigen",
  "meigenApiToken": "${MEIGEN_API_TOKEN}",
  "defaultModel": "nanobanana-2",
  "defaultAspectRatio": "16:9",
  "comfyuiUrl": "http://localhost:8188",
  "comfyuiDefaultWorkflow": "txt2img",
  "openaiApiKey": "${OPENAI_API_KEY}",
  "openaiBaseUrl": "https://api.openai.com/v1"
}

// AI agent will call this when user says:
// "find inspiration for product photography"
// "search for logo designs"
// "show me nature landscape prompts"

{
  "keywords": "product photography",
  "style": "commercial",
  "limit": 10
}

// After search_gallery, get full details
{
  "id": "12345"
}

// User says: "a coffee cup"
// AI calls enhance_prompt to expand into:
// "A ceramic coffee cup on a wooden table, warm morning light streaming through window, shallow depth of field, product photography style, commercial quality, 4K detail"

{
  "userPrompt": "a coffee cup",
  "style": "commercial",
  "aspectRatio": "16:9"
}

// Returns models from MeiGen Cloud, OpenAI, and local ComfyUI
{} // No parameters

// Basic generation
{
  "prompt": "a calico cat in a sunlit kitchen, photorealistic, 4K detail",
  "model": "nanobanana-2",
  "aspectRatio": "16:9"
}

// With reference image (local file auto-uploaded)
{
  "prompt": "product hero shot, professional lighting",
  "model": "gpt-image-2",
  "aspectRatio": "1:1",
  "referenceImagePath": "/Users/you/Desktop/bottle.jpg"
}

// With specific provider
{
  "prompt": "logo design, minimal, tech startup",
  "provider": "comfyui",
  "workflowName": "txt2img",
  "aspectRatio": "1:1"
}

// Text-to-video
{
  "prompt": "a cat walking through a garden, cinematic",
  "model": "seedance-2.0-fast",
  "duration": 5
}

// Image-to-video (first frame)
{
  "prompt": "camera slowly zooms in",
  "model": "happyhorse-1.0",
  "duration": 5,
  "firstFramePath": "/Users/you/Desktop/scene.jpg"
}

// Veo 3.1 (longest duration)
{
  "prompt": "product rotation, studio lighting",
  "model": "veo-3.1",
  "duration": 10
}

// List workflows
{
  "action": "list"
}

// View workflow details
{
  "action": "view",
  "name": "txt2img"
}

// Import new workflow (from ComfyUI API format JSON)
{
  "action": "import",
  "name": "my-flux-workflow",
  "workflow": { /* ComfyUI API format workflow */ }
}

// Delete workflow
{
  "action": "delete",
  "name": "old-workflow"
}

// Set preferences
{
  "action": "set",
  "preferences": {
    "defaultModel": "midjourney-v8.1",
    "defaultAspectRatio": "1:1",
    "favoriteStyle": "minimal modern",
    "favoritePrompts": ["tech logo", "product photography"]
  }
}

// Get preferences
{
  "action": "get"
}

undefined

undefined

undefined

undefined

undefined

undefined

-p, --prompt <text>          Image description (required)
-m, --model <name>           AI model (default: nanobanana-2)
-r, --aspect-ratio <ratio>   Aspect ratio: 1:1, 16:9, 9:16, etc. (default: 16:9)
--ref, --reference <path>    Reference image (local file, auto-uploaded)
--provider <name>            Force provider: meigen, openai, or comfyui
--workflow <name>            ComfyUI workflow name (when provider=comfyui)
--video                      Generate video instead of image
--duration <seconds>         Video duration: 5 or 10 (default: 5)
--first-frame <path>         First frame image for video (local file)
--no-wait                    Submit only, don't poll for completion
--json                       Machine-readable JSON output

// User says: "Create 4 product images for this perfume"
// AI will:
// 1. Call enhance_prompt 4 times to craft distinct prompts
// 2. Spawn image-generator sub-agents to run generate_image in parallel
// 3. Return all 4 images in ~2 minutes

// No special code needed — MeiGen handles orchestration automatically

// User drops ~/Desktop/bottle.jpg into chat and says:
// "Make this look more professional"

// AI calls:
{
  "prompt": "professional product photography, studio lighting, white background",
  "model": "gpt-image-2",
  "aspectRatio": "1:1",
  "referenceImagePath": "/Users/you/Desktop/bottle.jpg"
}

// MeiGen auto-compresses to <10MB and uploads to MeiGen Cloud

// 1. Export workflow from ComfyUI UI (Save API Format)
// 2. Import to MeiGen
{
  "action": "import",
  "name": "flux-dev-txt2img",
  "workflow": {
    "3": {
      "inputs": {
        "seed": 42,
        "steps": 20,
        "cfg": 7,
        "sampler_name": "euler",
        "scheduler": "normal",
        "denoise": 1,
        "model": ["4", 0],
        "positive": ["6", 0],
        "negative": ["7", 0],
        "latent_image": ["5", 0]
      },
      "class_type": "KSampler"
    },
    // ... rest of workflow
  }
}

// 3. Use in generate_image
{
  "prompt": "a cat",
  "provider": "comfyui",
  "workflowName": "flux-dev-txt2img"
}

// User gives minimal prompt: "a logo"
// AI automatically enhances before generating:

// Step 1: enhance_prompt
{
  "userPrompt": "a logo",
  "style": "minimal modern"
}
// Returns: "Minimal modern logo design, geometric shapes, bold typography, monochrome color palette, vector art style, professional brand identity, clean lines, scalable, white background"

// Step 2: generate_image with enhanced prompt
{
  "prompt": "Minimal modern logo design, geometric shapes...",
  "model": "midjourney-v8.1",
  "aspectRatio": "1:1"
}

// User: "Generate a logo with 3 different AI models"

// AI calls list_models to see what's available
{} // Returns: nanobanana-2, midjourney-v8.1, gpt-image-2, flux-2-klein, etc.

// Then generates with different models in parallel:
const models = ["nanobanana-2", "midjourney-v8.1", "flux-2-klein"];
for (const model of models) {
  generate_image({
    prompt: "Minimal tech startup logo, blue and white",
    model,
    aspectRatio: "1:1"
  });
}

/meigen:setup

export MEIGEN_API_TOKEN=meigen_sk_...

undefined

mcp_servers:
  meigen:
    timeout: 600          # 10 minutes
    connect_timeout: 120  # 2 minutes

npx meigen gen -p "..."

npm install -g meigen
meigen gen -p "..."

// In an AI coding agent's context, this is what happens when user says:
// "Create 3 logo variations for a tech startup"

import { spawn } from 'child_process';

const prompt = "Minimal tech startup logo, geometric, blue and white, modern";
const models = ["nanobanana-2", "midjourney-v8.1", "flux-2-klein"];

async function generateLogoVariations() {
  const jobs = models.map(model => {
    return new Promise((resolve, reject) => {
      const proc = spawn('npx', [
        'meigen', 'gen',
        '--prompt', prompt,
        '--model', model,
        '--aspect-ratio', '1:1',
        '--json'
      ]);

      let output = '';
      proc.stdout.on('data', (data) => { output += data; });
      proc.on('close', (code) => {
        if (code === 0) {
          resolve(JSON.parse(output));
        } else {
          reject(new Error(`Model ${model} failed`));
        }
      });
    });
  });

  const results = await Promise.all(jobs);
  results.forEach((result, i) => {
    console.log(`${models[i]}: ${result.imageUrls[0]}`);
  });
}

generateLogoVariations();

// Generate with custom ComfyUI workflow
import { exec } from 'child_process';
import { promisify } from 'util';

const execAsync = promisify(exec);

async function generateWithComfyUI() {
  const { stdout } = await execAsync(
    'npx meigen gen ' +
    '--prompt "portrait, cinematic lighting" ' +
    '--provider comfyui ' +
    '--workflow flux-dev-portrait ' +
    '--aspect-ratio 2:3 ' +
    '--json'
  );

  const result = JSON.parse(stdout);
  console.log('Image saved to:', result.localPath);
  return result.localPath;
}

#!/bin/bash

undefined

#!/usr/bin/env python3
import subprocess
import json
import os

def generate_video(prompt, first_frame=None, duration=5):
    cmd = [
        'npx', 'meigen', 'gen',
        '--prompt', prompt,
        '--video',
        '--model', 'seedance-2.0-fast',
        '--duration', str(duration),
        '--json'
    ]
    
    if first_frame:
        cmd.extend(['--first-frame', first_frame])
    
    result = subprocess.run(cmd, capture_output=True, text=True)
    data = json.loads(result.stdout)
    
    print(f"Video saved to: {data['localPath']}")
    return data['videoUrls'][0]

undefined

// ci-generate-assets.js
// Run in CI to generate marketing assets

const { execSync } = require('child_process');
const fs = require('fs');

const prompts = JSON.parse(fs.readFileSync('prompts.json', 'utf8'));

prompts.forEach((item) => {
  const cmd = `npx meigen gen --prompt "${item.prompt}" --model ${item.model} --aspect-ratio ${item.ratio} --no-wait --json`;
  
  try {
    const output = execSync(cmd, { encoding: 'utf8' });
    const result = JSON.parse(output);
    console.log(`Submitted: ${item.name} (ID: ${result.generationId})`);
  } catch (error) {
    console.error(`Failed: ${item.name}`, error.message);
    process.exit(1);
  }
});

console.log('All assets submitted. Check MeiGen dashboard for completion.');

{
  "3": { "class_type": "KSampler", "_meigen_role": "sampler" },
  "6": { "class_type": "CLIPTextEncode", "_meigen_role": "positive_prompt" },
  "7": { "class_type": "CLIPTextEncode", "_meigen_role": "negative_prompt" },
  "5": { "class_type": "EmptyLatentImage", "_meigen_role": "latent" }
}

{
  "provider": "comfyui",
  "comfyuiUrl": "http://localhost:8188",
  "meigenApiToken": "${MEIGEN_API_TOKEN}",
  "openaiApiKey": "${OPENAI_API_KEY}"
}

undefined

undefined