Content
# 🎨 MCP Image Gen
### Generate images with AI — locally, privately, no API keys.
> "Generate a cyberpunk city background for my video."
>
> Your AI assistant runs a model on your machine, saves the image, and can even import it straight into DaVinci Resolve.
[](#-why-this)
[](#-why-this)
[](#-three-models-pick-your-speed)
[](LICENSE)
---
## 🤔 What Is This?
An MCP server that lets AI assistants (Cursor, Claude, Windsurf) generate images using AI models running **entirely on your machine**. No cloud. No API keys. No subscriptions. No data leaves your computer.
Tell your AI assistant what you need, and it generates the image right there.
---
## ⚡ Why This?
Most image generation MCP servers are wrappers around cloud APIs — DALL-E, Midjourney, Gemini. You need API keys, you pay per image, and your prompts go to someone else's server.
**This one is different:**
- **Zero API keys** — nothing to sign up for, nothing to pay
- **Zero cloud** — models run locally via [OpenVINO](https://docs.openvino.ai/), optimized for CPU
- **Three quality tiers** — pick fast-and-rough or slow-and-detailed, depending on what you need
- **Pipeline-ready** — generated images work directly with DaVinci Resolve MCP and Video Editor MCP
Models download automatically from HuggingFace on first use and stay cached on your machine.
---
## 🎯 Three Models, Pick Your Speed
| | Model | Time | RAM | Resolution | Best for |
|---|-------|------|-----|-----------|----------|
| 🟢 **Fast** | SDXS-512 | ~20 sec | ~2 GB | 512×512 | Quick backgrounds, textures, gradients |
| 🔵 **Quality** | SDXL Turbo | ~30 sec | ~11 GB | 512×512 | Detailed assets, good prompt adherence |
| 🟣 **Premium** | SANA Sprint | ~2.5 min | ~13 GB | 1024×1024 | Highest quality — shines on GPU/NPU hardware |
All times are on CPU. If you have a GPU or NPU, the Premium tier gets dramatically faster.
---
## 🚀 Getting Started
### Step 1 — Set it up
```bash
git clone https://github.com/hiteshK03/mcp-image-gen.git
cd mcp-image-gen
bash setup.sh
```
### Step 2 — Tell your AI assistant about it
Add to `.cursor/mcp.json` (or equivalent):
```json
{
"mcpServers": {
"mcp-image-gen": {
"command": "/path/to/mcp-image-gen/venv/bin/python",
"args": ["/path/to/mcp-image-gen/server.py"]
}
}
}
```
### Step 3 — Ask for images
That's it. Your AI assistant can now generate images. Just ask:
```
"Generate a sunset gradient background and save it as bg.png"
"Create a dark tech UI mockup for my intro"
"Make a thank-you card with gold on black for my outro"
```
---
## 🔁 Part of a Video Production Pipeline
This server is designed to work alongside other MCP tools for a fully automated workflow:
```
You: "I need a cyberpunk background for the intro"
1. mcp-image-gen → generates the image locally
2. DaVinci Resolve MCP → imports it into the media pool
3. DaVinci Resolve MCP → places it on the timeline, scales to fit
All automatic. One conversation.
```
| Server | Role |
|--------|------|
| **This (mcp-image-gen)** | Generate images locally — backgrounds, textures, overlays, graphics |
| [**DaVinci Resolve MCP**](https://github.com/hiteshK03/davinci-resolve-mcp) | Control Resolve — import media, edit timelines, render, local AI |
| **Video Editor MCP** | File-based video processing — overlays, transitions, compositing |
The generated images are regular files. They work with `import_media` (Resolve), `add_image_overlay` (Video Editor), or anything else that accepts an image path.
---
## 🛠️ Tools
### `generate_image`
Generate an image from a text prompt.
| Parameter | Required | Description |
|-----------|----------|-------------|
| `prompt` | ✅ | What you want the image to look like |
| `output` | ✅ | Where to save it (PNG recommended) |
| `model` | | `sdxs-512`, `sdxl-turbo` (default), or `sana-sprint` |
| `width` | | Custom width in pixels |
| `height` | | Custom height in pixels |
| `seed` | | Fix the random seed for reproducible results |
### `list_image_models`
Returns all available models with specs, estimated times, and whether each is currently loaded in memory.
---
## 📋 Requirements
- Python 3.10+
- 2–13 GB RAM depending on which model you use
- No GPU required (but Premium tier benefits from one)
---
## 📄 License
MIT — use it however you want.
