OpenMAIC lib/media Image and Video Generation Providers: Complete 2024 Guide

OpenMAIC's lib/media module supports 11 image generation providers and 3 video generation providers through a unified adapter-based architecture.

The lib/media directory in the THU-MAIC/OpenMAIC repository provides a clean abstraction layer for image and video generation. Rather than calling third-party APIs directly, developers interact with provider-agnostic interfaces while the framework handles service-specific implementations.

Image Generation Providers in lib/media

OpenMAIC ships with 11 distinct image generation adapters, covering both commercial APIs and self-hosted solutions.

Commercial API Adapters

• OpenAI DALL·E — openai-image-adapter.ts
• Qwen (Alibaba) — qwen-image-adapter.ts
• MiniMax — minimax-image-adapter.ts
• Grok (xAI) — grok-image-adapter.ts
• Seedream — seedream-adapter.ts
• Seedance — seedance-adapter.ts
• Kling — kling-adapter.ts

Specialized Platform Adapters

• Nano-Banana — nano-banana-adapter.ts
• Lemonade — lemonade-image-adapter.ts
• HappyHorse — happyhorse-adapter.ts

Self-Hosted Adapter

• ComfyUI — comfyui-image-adapter.ts

The comfyui-image-adapter.ts enables local image generation by interfacing with ComfyUI workflows, making OpenMAIC suitable for privacy-sensitive deployments.

Video Generation Providers in lib/media

Video generation support is more focused with 3 specialized adapters:

Provider	Adapter File	Notes
VEO (Google)	veo-adapter.ts	High-quality text-to-video
MiniMax	minimax-video-adapter.ts	Same vendor, separate video API
Grok (xAI)	grok-video-adapter.ts	Video generation via xAI

Provider Registration Architecture

All adapters are registered in centralized registry files located at the lib/media root:

• image-providers.ts — Maps provider names like "openai-image" to their adapter classes
• video-providers.ts — Handles "veo-video", "minimax-video", "grok-video"

This registry pattern allows the higher-level media APIs to request generation without hardcoding service details. To add a new provider, implement the adapter interface and register the mapping in the appropriate providers file.

Example: Using a Provider Programmatically

While the raw source doesn't expose usage examples, the architecture implies this interaction pattern:

// Provider selection through registry lookup
const provider = mediaRegistry.getProvider('openai-image');
const result = await provider.generate({
  prompt: 'A futuristic robot in a cyberpunk city',
  size: '1024x1024'
});

Each adapter normalizes response formats so calling code receives consistent output regardless of underlying service.

Summary

• 11 image providers including DALL·E, Qwen, ComfyUI, and 7 others
• 3 video providers: VEO, MiniMax, and Grok
• All adapters live in lib/media/adapters/ with descriptive filenames
• Registration happens in image-providers.ts and video-providers.ts
• ComfyUI support enables fully self-hosted image generation
• The adapter pattern isolates service-specific logic from application code

Frequently Asked Questions

How do I add a custom image provider to OpenMAIC?

Create a new file in lib/media/adapters/ implementing the adapter interface, then import and register it in image-providers.ts with a unique provider key. The existing openai-image-adapter.ts serves as the canonical reference implementation.

Does OpenMAIC support local image generation without cloud APIs?

Yes. The comfyui-image-adapter.ts enables local generation by connecting to a self-hosted ComfyUI instance. This is the only fully offline-capable adapter in the current lib/media distribution.

Why are there separate adapters for MiniMax image and video?

MiniMax operates distinct API endpoints and authentication flows for image versus video generation. The minimax-image-adapter.ts and minimax-video-adapter.ts each handle service-specific request/response formatting while exposing a unified interface to the registry.

What determines which provider is used at runtime?

The provider registry in image-providers.ts or video-providers.ts maps string keys to adapter instances. Application configuration or user selection specifies the provider name, and the registry resolves the appropriate adapter implementation transparently.

{
  "mcpServers": {
    "instagit": {
      "command": "npx",
      "args": ["-y", "instagit@latest"]
    }
  }
}