Can you use it in client work? What the open image model licences actually say

"Open weights" does not mean "yours to sell work from". Across the two biggest open image model families, some weights are Apache 2.0 and genuinely unrestricted, some are licensed for non-commercial use only, and the clause everyone quotes as proof they can bill for the output does not say what it looks like it says.

Here is the state of it, taken from the licence files themselves on 20 September 2026. This is a designer's reading of published terms, not legal advice, and licences get revised: check the current file before you invoice anything.

The short version

The pattern is consistent and easy to remember: in the FLUX families, the small distilled model is the open one and everything better is not. The Qwen team puts the standard Apache 2.0 text in its Qwen-Image repository, which is the most permissive position any major image model maker currently takes. The weights themselves are distributed on Hugging Face and ModelScope, so read the licence on the specific checkpoint before you build on it.

The clause everyone misreads

Both FLUX non-commercial licences contain this, in the section on Outputs:

We claim no ownership rights in and to the Outputs. You are solely responsible for the Outputs you generate and their subsequent uses in accordance with this License. You may use Output for any purpose (including for commercial purposes), except as expressly prohibited herein.

Read on its own, that looks like permission to use FLUX.1 [dev] pictures in paid work. It is not, because of two things around it.

First, the grant itself. The licence permits you to access and use the model "solely for your Non-Commercial Purposes", and the definition of that term excludes revenue-generating activity, anything connected to your employment responsibilities, and any use that has an impact on end users. Making the image for a client is already outside the grant, whatever happens to the file afterwards.

Second, the restrictions section prohibits use of "the FLUX Model (or any Derivative thereof, or any data produced by the FLUX Model)" for any commercial or production purposes. Data produced by the model is the output. That is the "except as expressly prohibited herein" that the Outputs clause points at.

So the accurate summary is narrower than the internet's: Black Forest Labs does not claim to own your pictures, and under the non-commercial licence you still may not make them for money. If you want to, the licence tells you where to ask, and their FLUX.1 repository notes that commercial licensing is priced on monthly usage, with a --track_usage flag in their own inference code for reporting it.

The obligations even the free-to-use path carries

Two requirements in the non-commercial licences are easy to miss because they read like boilerplate and are not.

Filter or review before anything goes out. You have to either run content filtering for unlawful or infringing content, or review output for it before distribution, which the licence defines broadly enough to include non-public sharing. Sending a client a contact sheet counts.

Disclose AI generation where the law requires it. The licence asks you to ensure output carries a disclosure that it was generated or modified with AI, to the extent applicable law requires. It pushes the compliance question down to you and your jurisdiction, and the licence itself cites the EU AI Act, GDPR and the CCPA in its restrictions.

The restrictions list is also worth reading once as a working designer rather than as a lawyer. It rules out biometric processing and surveillance work, and it names rights of publicity and "digital replica" rights explicitly. If your brief involves a real person's face or voice, that clause is about you.

What Apache 2.0 actually gets you

For FLUX.1 [schnell], FLUX.2 [klein] 4B and the Qwen-Image line, the answer is the boring, good one. Apache 2.0 lets you use, modify and distribute the weights commercially, including inside a product, as long as you keep the licence and notices with any copies you distribute, and it includes an express patent grant. It puts no conditions at all on the images you make.

That makes the practical decision for a studio simple. If work is going to be billed, build the pipeline on the Apache 2.0 models or on a hosted API whose terms allow it, and keep the non-commercial weights for personal work, tests and pitches you are not charging for. Swapping model mid-project because a licence surfaced late is far more expensive than starting on the permissive one.

What a licence still does not settle

None of this touches the two questions clients ask most.

Who owns the image? The model makers here disclaim ownership of outputs, which is not the same as granting you copyright. Whether a generated image is protectable at all, and by whom, is a question of national law and is being answered differently in different places. Put it in the contract rather than assuming it.

Is the image safe to publish? A licence cannot tell you whether your output resembles someone's trademark, a photographer's frame or a living person. Both FLUX licences put that squarely on you, which is another way of saying the review step is not optional.

If you are choosing between these models on capability rather than terms, that comparison is in which AI image model for which job.

Sources

• FLUX.1 [dev] Non-Commercial License v1.1.1, read 20 September 2026
• FLUX Non-Commercial License v2.1, read 20 September 2026
• Black Forest Labs, FLUX.1 repository, for the model and licence table and commercial usage tracking
• Black Forest Labs, FLUX.2 repository, for the FLUX.2 model and licence table
• Qwen-Image repository licence file, Apache License 2.0