The M-series chip in your MacBook shares one pool of memory between CPU and GPU, so a 16GB Mac can hand most of that to an image model with no copying across a bus. That unified memory is exactly what Stable Diffusion wants, and it sits idle while you pay a monthly fee to generate images on a rented server. Off Grid AI Desktop is a free, open-source app that generates images directly on your Mac.
GitHub → Free, open source, runs offline.
It bundles stable-diffusion.cpp and drives it with Apple's Metal backend. Your prompts and your images never leave the machine.
What You Need
Apple Silicon is the requirement. Metal acceleration and unified memory are what make this practical.
| Tier | Chip / RAM | What runs well |
|---|---|---|
| Minimum | M1, 8GB | SD 1.5, SDXL-Lightning at low steps |
| Recommended | M2 or M3, 16GB | Full SDXL, Z-Image-Turbo, comfortable |
| Comfortable | M3 Pro/Max or M4, 32GB+ | Large SDXL batches, fast iteration |
You need macOS recent enough for Metal (Monterey or later) and roughly 6-12GB of free disk per image model. Models download on demand, so you only pull what you use.
What Off Grid AI Desktop Can Do
You get a full local image studio. Type a prompt, watch it paint, keep the result. No browser tab, no queue, no credit counter.
Run the models that matter in 2026. SDXL for quality. SDXL-Lightning for fast few-step output. SD 1.5 and 2.1 when you want something light. Z-Image-Turbo, the 2026 flagship, reaches a finished image in about 8 steps.
Watch it generate. A live per-step preview shows the image forming. You see progress and an ETA, and you can cancel a run the moment you know it is going wrong.
Start from text or from an image. Text-to-image builds from a prompt. Image-to-image takes a picture you drop in and reworks it toward your prompt, keeping composition you like.
Skip the prompt engineering. Style presets like Sketch, Cinematic, and Anime apply a tuned look so you describe the subject and let the preset handle the rest.
Keep your work. Every image lands in an artifacts gallery. Click any one to open it full size in a lightbox.
Which Models to Use
Match the model to your memory and your patience.
| Your Mac | Model | What to expect |
|---|---|---|
| M1 / 8GB | SD 1.5, SDXL-Lightning | Quick results, modest resolution |
| M2-M3 / 16GB | Z-Image-Turbo, SDXL-Lightning | Strong images in a handful of steps |
| M3 Pro+ / 32GB | Full SDXL | Best quality, larger batches |
SDXL-Lightning and Z-Image-Turbo are the comfortable defaults on most Macs. They reach a usable image in far fewer steps than standard SDXL, which is the single biggest lever on how long you wait.
How Hardware Acceleration Works
Image generation is a denoising loop. The model starts from noise and refines it step by step, and each step is dense matrix math. Metal runs that math on your Mac's GPU cores directly.
Unified memory is the Apple-specific advantage. On a PC with a discrete GPU, model weights copy from system RAM into separate VRAM. On Apple Silicon there is one pool, so the GPU reads the weights in place. That is why a 16GB Mac handles SDXL more gracefully than the raw number suggests.
The models are quantized so they fit. Reducing weights from 16-bit to 8-bit or 4-bit shrinks an SDXL checkpoint enough to sit alongside everything else you have open.
Getting Sharper Images Faster
A few settings move the needle most.
Steps drive both quality and time. With Lightning or Turbo models, stay near their intended low step count. Adding steps to a few-step model wastes time without improving the result.
Resolution costs quadratically. Doubling each side quadruples the work. Generate at the model's native size, then upscale if you need more.
Use img2img to refine. Rather than re-rolling text-to-image and hoping, take a result you half-like and push it with a low-strength img2img pass.
Privacy: Stronger Than a Cloud Image Generator
A cloud image service uploads your prompt and stores your output. Many reserve the right to train on what you make. Some require a paid plan before you can generate anything at all.
Off Grid AI Desktop does the work on your GPU. Nothing uploads. There is no account, no telemetry, and no credit meter. The code is AGPL-3.0, so the privacy claim is something you can check rather than trust. Turn off Wi-Fi and it still generates.
Getting Started
- Open the repo: github.com/off-grid-ai/desktop.
- Download the macOS build, or clone and build from source.
- Launch the app. It uses Metal automatically on Apple Silicon.
- Open image generation, pick a model that fits your RAM, and download it.
- Type a prompt, pick a style preset, and generate. Your images stay on your Mac.
What's Coming
- More image models as open weights land.
- Cross-device sync, so your artifacts gallery follows you, still private.
- Tighter integration between image generation and the rest of the local studio.
Grounded roadmap. The local spine comes first.
FAQ
Q: Is it really free?
Yes. The app is free and open source under AGPL-3.0. No subscription, no account, no per-image charge.
Q: Does it work offline?
Yes. Once a model is downloaded, you can disconnect entirely. Generation is local.
Q: Which image models can I run?
SDXL, SDXL-Lightning, SD 1.5, SD 2.1, and Z-Image-Turbo. Pick and download them from inside the app.
Q: How much RAM do I need?
8GB on an M1 runs SD 1.5 and Lightning models. 16GB makes full SDXL and Z-Image-Turbo comfortable.
Q: Does it support image-to-image?
Yes. Drop in an image and steer it with a prompt. There are also style presets and a live per-step preview.
Q: Is my data private?
Yes. Prompts and images never leave your Mac. No telemetry, no account, and the source is open.
Generate images on the Mac you already own. github.com/off-grid-ai/desktop
