Free AI Image Generation at Home: Stable Diffusion and FLUX

You can generate unlimited AI images free on your own computer using Stable Diffusion or FLUX, with no subscription and no per-image credits. The most valuable thing to understand before you start is the licensing: FLUX.1 dev is labelled "non-commercial," but its own licence explicitly permits selling the images you make with it. That single distinction decides whether you can build a business on free tools.

Key takeaways

• FLUX.1 dev's outputs are commercially usable even though the model is non-commercial. Black Forest Labs' licence states you may use output "for any purpose (including for commercial purposes)."
• FLUX.1 schnell is Apache-2.0, making both model and output fully commercial. It is the cleanest option for sellers.
• AUTOMATIC1111, the tool most guides still recommend, has not had a release since February 2025 and cannot run FLUX. Use ComfyUI, Fooocus or InvokeAI instead.
• Video memory is the real constraint: SD 1.5 runs on 4GB, SDXL wants 8-12GB, FLUX.1 dev typically needs 16GB.
• Stability's SD3.5 licence allows free commercial use only under $1 million in total organisational revenue, not revenue from the model.

What running Stable Diffusion locally means

Stable Diffusion and FLUX are AI image models you download and run on your own graphics card. Unlike Midjourney or DALL-E, there is no subscription, no credit meter and no per-image cost. Once the model is on your drive, you can generate continuously.

You need two things: the model itself, and an interface to run it. The interface choice matters more than most guides suggest, because the most-recommended option is no longer maintained.

Which interface to use in 2026

Search for a beginner guide and it will almost certainly tell you to install AUTOMATIC1111. That advice is out of date.

AUTOMATIC1111 has 164,800 stars and remains the most-referenced interface, but its last release was version 1.10.1 in February 2025. More importantly for anyone starting now, it cannot run FLUX.1, the current quality leader. Recommending it in 2026 sends beginners to a dead end.

ComfyUI (131,100 stars, GPL-3.0) is the active choice. It ships weekly releases, has an official desktop app for Windows and macOS, and supports everything current: SD 1.5, SDXL, SD3.5, FLUX.1, FLUX.2 and Qwen Image. Its node-based interface is genuinely harder at first, but it is where development is happening.

Fooocus (52,700 stars, GPL-3.0) is the friendliest option, designed to work like Midjourney with minimal configuration. It states a minimum of 4GB Nvidia video memory and 8GB system RAM. Note that it has moved to "limited long-term support, bug fixes only," so it is stable rather than growing.

InvokeAI (28,100 stars, Apache-2.0) sits in between, with a proper launcher installer aimed at non-technical users and a more conventional interface.

For a first install, Fooocus is the gentlest start and ComfyUI is the one worth learning.

What hardware you actually need

Video memory, not system RAM, is the binding constraint.

On Macs, Apple Silicon uses unified memory rather than dedicated VRAM. SDXL wants an M-series machine with at least 16GB, and FLUX in reduced precision is comfortable on an M4 Pro with 24GB. Intel Macs are effectively out.

If your hardware falls short, quantised model versions trade some quality for a much smaller memory footprint, which is how FLUX drops from 16GB to around 7GB.

The licensing question that actually matters

This is the section worth reading twice, because getting it wrong either costs you money or stops you using the best free model available.

FLUX.1 dev is described as non-commercial, and most articles stop there and tell readers they cannot sell FLUX images. That is wrong. The licence from Black Forest Labs states directly:

"We claim no ownership rights in and to the Outputs... You may use Output for any purpose (including for commercial purposes), except as expressly prohibited herein."

The non-commercial restriction applies to the model, not the images. You may not build a paid service or product around FLUX.1 dev itself, and you may not use its output to train a competing model. But selling prints, using images in client work or putting them on products is permitted.

FLUX.1 schnell is Apache-2.0, so both model and output are fully commercial with no ambiguity. If you want zero licensing risk, this is the answer.

Stable Diffusion 1.5 and SDXL use the CreativeML Open RAIL-M licence, which permits commercial use royalty-free. Its exact wording: "Except as set forth herein, Licensor claims no rights in the Output You generate using the Model." It attaches use-based restrictions prohibiting illegal content, harassment, discrimination and unlicensed professional advice.

SD 3.5 uses Stability's Community License, which permits free commercial use only while your organisation earns under $1 million a year in total revenue, explicitly "regardless of whether derived from Stability's Models." Most articles misstate this as revenue from AI work. Above that threshold you need an enterprise licence. Stability does state it "will never ask you to delete resulting images."

Where the copyright litigation stands

Anyone selling AI art should know the legal position, which shifted meaningfully in late 2025.

In Getty Images v Stability AI, decided in the English High Court in November 2025, Getty dropped its main training and output copyright claims mid-trial, and its secondary infringement claim was rejected. The court accepted that model weights do not contain copies of the training images. Only limited trade mark findings survived.

In the United States, Andersen v. Stability AI in the Northern District of California is further along: DMCA claims were dismissed with prejudice in 2024, but direct copyright claims are proceeding, and trial was scheduled for 8 September 2026. That is imminent as of writing, and an outcome could change the picture, so check the current status before relying on this.

The practical position today is that generating and selling images from these models is legal and widely done. The unresolved questions concern how the models were trained, not what you make with them.

Who should generate images locally

Do it if you have a graphics card with 8GB or more, you generate enough images that credits are a real cost, you want full control over models and settings, or you need images to stay on your machine.

Stick with paid tools if your hardware is limited, you value polish over control, or you generate occasionally enough that a subscription is cheaper than the time spent setting up. The free tiers of cloud tools are also genuinely capable for light use.

For a direct comparison, see free local image generation versus Midjourney and DALL-E. To compare the free options against each other, read the best free AI image generators.

The verdict

Running image models locally is the best value in AI right now: unlimited generation, no subscription, and with FLUX.1 schnell or SD 1.5, clean commercial rights.

The single most useful thing to take away is the output-versus-model distinction. FLUX.1 dev produces the best free images available and its licence explicitly permits selling them. A lot of people are avoiding it because they read the words "non-commercial" and stopped. Read the actual clause, and the best free image model is available to you.

Licensing is the part most guides get wrong. Our guide to selling AI art and model licences explains why FLUX dev output is commercially usable.

Related reading: free alternatives to Midjourney and how to sell AI art on Etsy.

Pricing verified against each vendor's own pricing page in September 2026. Plans change often, so check the vendor's page before you buy.