Draw Things: Local AI Image Generator

Draw Things brings full-featured AI image generation to Apple devices - for free, offline, and with zero setup beyond installing the app.

At a Glance

TL;DR - Is It Worth It?

Yes, if you own an Apple device. Draw Things is one of the most full-featured ways to run AI image generation locally on Mac, iPhone, or iPad. It's free, it's private, and it gets weekly updates with support for the latest models. If you're on Windows or Linux, you can't use it - look at Fooocus or ComfyUI instead.

Top 5 Features

1. 100% offline generation - Once you download a model, everything runs on your device. No internet needed. Your prompts and images never leave your phone or computer.

2. Native Apple Silicon optimization - The app is written in Swift (Apple's programming language) and uses Metal FlashAttention, a technology that squeezes maximum speed from M-series chips. The developer's documentation credits Metal FlashAttention with roughly 20% faster generation on M3 and M4 chips.

3. Huge model library - Draw Things supports Stable Diffusion 1.5, SDXL, FLUX.1, HiDream, Qwen Image, and more. You can download models right inside the app or import your own from sites like Hugging Face and CivitAI.

4. On-device LoRA training - LoRA (Low-Rank Adaptation) is a way to teach an AI model new concepts - like your face, your art style, or a specific character. Draw Things lets you do this training directly on your device.

5. Full creative toolkit - Text-to-image, image-to-image, inpainting (editing parts of an image), outpainting (extending an image beyond its borders), ControlNet (guiding the AI with poses or edges), infinite canvas, scribble-to-image, and even AI video generation. That's a lot packed into one free app.

Requirements & Setup

Setup is about as easy as it gets. You download Draw Things from the App Store, open it, pick a model to download, and you're generating images. No terminal commands, no Python, no Docker. The whole process takes about 5 minutes.

If you're on an older or lower-RAM device (like an iPhone 11), you can still use it - the app has a "Server Offload" feature that lets you send the heavy computation to a more powerful Mac on your home network.

Which model should you use in Draw Things?

Pick Stable Diffusion 1.5 for fast drafts, SDXL for general quality, and FLUX.1 for current-generation detail; HiDream and Qwen Image are further options in the app once you know what you want. All of them need at least 8 GB of RAM for images, and 16-32 GB is the comfortable range.

Sizes are the ones quoted in the Limitations section below. We don't list speed timings for these models, because we haven't published a measured test.

NSFW models and what Draw Things allows

Yes, Draw Things allows NSFW content: the Details row lists "NSFW Allowed: Yes", and the app can import community models from CivitAI and Hugging Face. This section is for adults only (18+) and stays factual. How you use it is your responsibility, including following the rules of the site you download from and your local laws.

To pick and import a community model:

1. Choose the base model first. Community models are built on a base such as SD 1.5, SDXL, or FLUX.1. Choose one that matches your device's RAM and free storage (see the table above).
2. Read the model page. Check the base model, the file format, the license or usage terms, and any settings the author recommends.
3. Check the file type. Draw Things imports .safetensors and .ckpt checkpoints, plus LoRAs and embeddings.
4. Import it. Follow the steps in the next section.
5. Start with the author's settings. Use the recommended steps and sampler as a baseline, then change one thing at a time.

Import a model from CivitAI or Hugging Face

You can bring your own models into Draw Things instead of using only the ones the app offers. Supported formats are .safetensors and .ckpt files, LoRAs, and embeddings.

1. Open the model's page on CivitAI or Hugging Face and confirm its base model and file format.
2. Copy the model's download URL. On CivitAI model pages, you can also use the Play button.
3. Paste the URL into Draw Things' model import.
4. Wait for the download to finish. Large models take a while, and an SDXL model is 6-7 GB.
5. Select the imported model and generate a test image before you rely on it.

Limitations

• Apple-only - No Windows, Android, or Linux version exists. The developer has said Android would require too many physical devices to test with a one-person team. If you're not on Apple hardware, check out Easy Diffusion or ComfyUI.

• Not open source - You can't view or modify the source code. The community-scripts repository on GitHub is open, but the core app isn't. If source-code transparency matters to you, ComfyUI or AUTOMATIC1111 are fully open.

• Storage-hungry - AI models are big files. A single SDXL model is 6-7 GB. FLUX.1 models can be 12+ GB. If you download several models plus keep your generated images, you'll burn through storage fast. Budget at least 50 GB of free space for serious use.

• UI has rough edges - The interface works, but the settings panel requires a lot of scrolling. Browsing your image history could be smoother. It's functional, not polished. The developer prioritizes new model support and performance over UI polish.

How It Compares

For a broader look at what's available across all platforms, tools like LocalForge AI bundle Stable Diffusion with pre-configured models and extensions - a good option if you want zero setup on Windows too. Easy Diffusion is another cross-platform alternative worth checking.

Bottom Line

Who should use Draw Things:

• Apple users who want the easiest path to local AI images - App Store install, no terminal, works in 5 minutes.
• Privacy-conscious creators - Everything stays on your device. No cloud, no accounts, no data collection.
• iPhone/iPad owners - It's the only serious option for running full Stable Diffusion models on mobile Apple hardware.

Who should skip it:

• Windows or Linux users - It doesn't exist on your platform. Use ComfyUI, Fooocus, or Easy Diffusion instead.
• Power users who need node-based workflows - ComfyUI gives you more granular control over every step of the pipeline.
• Users with less than 8 GB RAM - You'll hit crashes or extremely slow generation. The Server Offload feature can help, but only if you have a more powerful Mac on the same network.