An open source image generator usually comes in two parts with two licences: the interface or library you install, and the model weights that make the images. The interface can be fully open while the weights limit commercial use, so check both before you make images for a business.
Most people who download an open source image generator read one licence and miss the other. The app is open source, so they assume the pictures are free to use however they like. That isn't always true, because the images come from a model, and the model has its own terms.
So this page keeps the two apart. Below you'll find interfaces, a plugin, a library and two model families, each with its code licence and its weights licence side by side. They're listed in no particular order.
Two licences on every picture
Think of it as the camera and the film. The interface is the camera: ComfyUI, InvokeAI, SD.Next, SwarmUI and the painting-app plugin are open source tools you run on your own machine. But the images come from model weights you download separately, and those weights carry their own licence.
Even one model family can split. On Hugging Face, the FLUX.1 schnell model card lists Apache-2.0, while the FLUX.1 dev card lists a non-commercial licence. Stable Diffusion 3.5 uses Stability's own community licence, which is a custom licence, not a standard open source one. That's why these are better called open weights than open source. If you're making marketing images for a business, the weights licence is the one that decides what you can do, so read the model card, not just the interface's licence file.
What you're installing
Workflow engines and web interfaces
These run in your browser against a server on your own computer. They range from a node graph you wire up yourself to a one-click installer; Easy Diffusion sits at the beginner end.
A plugin and a library
Krita AI Diffusion adds generation to a painting app you already use. Hugging Face Diffusers is for developers who'd rather write a few lines of Python than use an interface.
The model families
These are the weights themselves. You load them into one of the tools above; on their own, they aren't something you click on.
Code licence and weights licence for each
| Tool | Code licence | Weights licence |
|---|---|---|
| ComfyUI | GPL-3.0 | Set by whichever model you load |
| Easy Diffusion | Custom: MIT-style Section I | Section II points to CreativeML Open RAIL-M |
| FLUX.1 (Black Forest Labs) | Apache-2.0 inference code | Schnell Apache-2.0; dev non-commercial licence |
| Hugging Face Diffusers | Apache-2.0 | Each model on the Hub sets its own |
| InvokeAI | Apache-2.0 | Set by whichever model you load |
| Krita AI Diffusion | GPL-3.0 | Set by the models you use |
| SD.Next | Apache-2.0 | Set by whichever model you load |
| Stable Diffusion 3.5 (Stability AI) | Not applicable: a model release | Stability's own community licence |
| SwarmUI | MIT | Set by whichever model you load |
Nine tools and models, in no particular order
ComfyUI
Best for: Building image workflows as a graph of nodes on your own machine, with a desktop app and a hosted option from the same company too.
Easy Diffusion
Best for: Beginners wanting a one-click installer and a browser interface for making images on their own PC, as its repo describes it.
FLUX.1 (Black Forest Labs)
Best for: Running the FLUX.1 text-to-image models with official inference code, choosing between schnell and dev weights published on Hugging Face.
Hugging Face Diffusers
Best for: Python developers running or training diffusion pipelines in a few lines of code. It's a code library, not a point-and-click app.
InvokeAI
Best for: Artists who want a locally hosted web server with a canvas for generating, refining and iterating on images, under what its README calls a commercially-friendly licence.
Krita AI Diffusion
Best for: Painters already in Krita who want generation, inpainting and outpainting inside the app, with their own models on their own hardware or cloud generation.
SD.Next
Best for: Mixed hardware: its README says it supports many diffusion models on NVIDIA, AMD, Intel, Apple Silicon or plain CPU, for images and video.
Want AI doing the heavy lifting in your marketing?
I build the systems that handle the boring 80 percent, so you get your week back. Done properly, with the human kept in.
Stable Diffusion 3.5 (Stability AI)
Best for: Downloading open text-to-image weights to run in an interface of your choice, under a custom community licence rather than a standard open source one.
SwarmUI
Best for: Running local models through a modular web interface; note its latest tagged release is still marked Beta.
My advice on picking a set-up
If you've never run an image model before, start with the most beginner-friendly interface and one model whose card plainly allows the use you have in mind. Once you know what you want to make, move to a node-based engine or a canvas tool for more control. If you're a developer adding generation to a product, the Python library gives you the cleanest route in, and the licence question gets more important, not less.
For anything used commercially, keep a note of which model and version made each image and what its card said at the time. It's a small habit that saves an awkward conversation later. If you'd rather not run anything locally, there are hosted options in our guides to AI photoshoot generators and AI image upscalers.
The hardware question to answer before you download
Running these tools locally means your computer does the work. A recent graphics card makes generation far quicker, though some interfaces also support Apple Silicon or even a plain CPU, slowly. Models are large downloads, so leave plenty of disk space, and expect to try a few settings before images come out the way you want. If your machine struggles, several projects here offer a hosted or cloud option from the same makers.
Do you make an open source image generation tool or model? Tool makers can ask to be considered for a featured place on this list.
Want your tool on this list? Featured spots and reviews here are paid partnerships. Here's how to partner with me.
Frequently asked questions
Can I use images from an open source image generator commercially?
It depends on the model, not just the app. Some model cards list a permissive licence, others a non-commercial or custom community licence. Read the card for the exact model you use.
What's the difference between open source and open weights?
Open source usually means the code is under an approved open licence. Open weights means the trained model files are downloadable, but they may carry a custom licence with conditions.
Do I need a powerful computer to run AI image generation locally?
A recent graphics card helps a lot. Some tools also run on Apple Silicon or a CPU, but generation is much slower without a capable graphics card.
Are these tools free to use?
The interfaces and libraries here are free to download as of Nov 2026. Some makers also run hosted versions, and the model you load still sets its own terms.
Which interface is easiest for beginners?
A one-click installer with a browser interface is the gentlest start. Node-based engines give more control but take longer to learn.
Do the model weights come with the interface?
Usually not. You download model weights separately, often from Hugging Face, and load them into the interface you've installed.
