Short answer to “can you use Civitai models commercially”: often yes, but Civitai is not a license, just the site that hosts the files. Every checkpoint, LoRA or merge you download there carries two layers of terms: the permissions its author chose when uploading it, and the license of the base model it was trained on, such as SDXL, FLUX.1 [dev] or WAN 2.2. Commercial use is only an option when both layers allow it, and when what you generate doesn’t use someone else’s brand, character or face.
Below is what we check before a client’s ComfyUI template goes into paid production: where to look on a Civitai model page, what the main base model licenses say, why selling images and running a paid service are different things, and a checklist you can run yourself.
This is not legal advice. It summarizes public license texts as we read them. Licenses and Civitai’s interface change, so check the current text before you rely on it. Sources were checked in October 2026.
Where Civitai shows the license and permissions
Everything you need is in the details block of the model page, next to the version selector:
- License line. The base model license for the selected version, for example
CreativeML Open RAIL++-Mfor SDXL checkpoints orFLUX.1 [dev] Non-Commercial Licensefor Flux LoRAs. One page can hold versions for different base models, so check the version you actually download. - Permission icons. Small icons, green for allowed and red for restricted, with tooltips such as “No commercial use” or “No generation services”. Clicking them opens a list titled “This model permits users to:” with a check or a cross next to each item.
- Addendum. For SD 1.5 and SDXL 1.0 models, both on Open RAIL licenses, Civitai turns the author’s restrictions into a license document and links it as “Addendum”. The restrictions appear there as plain clauses: creator credit, generation services, sale of the model, sale of merges.
- Description. Authors often add their own terms or a commercial licensing contact here.
The items in that list, in Civitai’s wording:
| Item | What it tells you |
|---|---|
| Use the model without crediting the creator | If crossed out, the author requires credit; the addendum shows the format |
| Sell images they generate | The key item if you sell or publish the outputs |
| Run on services that generate for money | Paid generation services outside Civitai |
| Run on Civitai | Civitai’s own on-site generator |
| Share merges using this model | Merging or fine-tuning on top of it |
| Sell this model | Selling or licensing the weights themselves |
| Sell merges using this model | Selling merges that include it |
| Have different permissions when sharing merges | Whether merges may carry different terms |
Two caveats. The author sets these checkboxes at upload, the commercial ones are on by default, and they can be edited later, so a green icon may just mean untouched defaults; save a dated screenshot. And Civitai generates the addendum only for SD 1.5 and SDXL 1.0 models: for other base models the checkboxes are not turned into license text, so read them as the author’s stated terms; they do not replace the base model license.
The base model license applies to every LoRA and merge
A LoRA or a merged checkpoint is a derivative of its base model, and the base license comes with it. Civitai permissions can narrow what you may do; they cannot grant rights the base license doesn’t give. The FLUX.1 [dev] license says directly that its restrictions apply to any derivative, and CreativeML Open RAIL++-M requires its use restrictions to be carried into the terms of any derivative. Here is what the most common base models say at the source.
SDXL: CreativeML Open RAIL++-M
SDXL 1.0 base and refiner are under CreativeML Open RAIL++-M. It grants a royalty-free license with no non-commercial clause, explicitly allows hosting the model for third parties, and the licensor claims no rights in your output while holding you accountable for it. In exchange you accept use-based restrictions: nothing unlawful, nothing that harms minors, no defamation or harassment, no harmful disinformation, no medical advice, and a few more. Anyone you pass a derivative to must receive the same restrictions.
Not every model with SDXL in its name uses this license: SDXL Turbo is under the Stability AI Community License, with its own registration and revenue-threshold conditions for commercial use.
FLUX.1 [dev]: non-commercial license, outputs treated separately
FLUX.1 [dev] is under the FLUX.1 [dev] Non-Commercial License (version 1.1.1 when we checked); on Hugging Face, Kontext [dev] and Krea [dev] list the same license. The text pulls in two directions:
- The model itself may only be used for non-commercial purposes. The definition explicitly excludes revenue-generating activity and use that affects end users, and the restrictions section rules out commercial or production purposes. Anything else needs a separate license from Black Forest Labs.
- The outputs are treated separately. BFL claims no ownership of them and allows using them for any purpose, including commercial, except where the license prohibits it. You may not use outputs to train a model that competes with FLUX.1 [dev] or Kontext [dev], and you need either content filtering or a review of outputs for unlawful or infringing content, plus an AI disclosure where the law requires one.
So “can I sell images made with Flux” has no one-word answer. The outputs clause allows commercial use of images; the model clause limits how you may run the model, and paid production for clients looks like exactly the revenue-generating use it excludes. The cautious route for paid production is a commercial license from BFL or a different base model. In October 2026, BFL’s licensing page covers its newer FLUX.2 and FLUX 3 models and doesn’t mention FLUX.1 [dev], so terms for it have to be requested from BFL directly. FLUX.2 [dev] is also published under a non-commercial license.
FLUX.1 [schnell]: Apache 2.0
FLUX.1 [schnell] is under Apache 2.0, and its model card says it can be used for personal, scientific and commercial purposes. It is a different model, distilled for 1 to 4 steps, so moving a [dev] template to schnell means re-testing settings and LoRAs, not swapping a filename. Still, it is the first alternative to try inside the Flux family.
WAN 2.2: Apache 2.0
The WAN 2.2 models published by Wan-AI on Hugging Face, including the image-to-video I2V-A14B, are under Apache 2.0. The model card adds that the authors claim no rights over your generated content and that you are fully accountable for its use. In a WAN template the base model is rarely the issue; the LoRAs, the starting image and the content are.
| Base model | License | Commercial use of the model | Outputs |
|---|---|---|---|
| SDXL 1.0 | CreativeML Open RAIL++-M | Allowed, with use restrictions | Licensor claims no rights; you are accountable |
| FLUX.1 [dev] | FLUX.1 [dev] Non-Commercial License | Only with a separate license from BFL | Commercial use allowed except where prohibited; conditions apply |
| FLUX.1 [schnell] | Apache 2.0 | Allowed | Model card: personal, scientific and commercial use |
| WAN 2.2 | Apache 2.0 | Allowed | Authors claim no rights; you are accountable |
Selling images is not the same as running a paid service
Civitai splits commercial use into separate permissions on purpose. “Sell images they generate” is about outputs: you generate, then sell or publish the result. “Run on services that generate for money” is about offering the model itself to people who pay for generation. In the Civitai addendum this becomes a ban on using the model on any service that monetizes generation, with subscription fees, per-image fees and ad-supported generators as examples.
Base licenses draw their own lines. Open RAIL++-M explicitly allows hosting for third parties, as long as the use restrictions go with it. Read literally, the FLUX.1 [dev] license covers neither paid production nor a paid service; for those it points to a separate license from BFL. Apache 2.0 has no field-of-use restriction.
Done-for-you generation sits between the two. When a studio runs your template and is paid per image or clip, you receive the outputs, but the studio is also paid to run the model. That is why for contract work we look at both permissions, not only “Sell images”, and flag a closed one before the start.
Brands, real people and other people’s characters
A model license covers the weights, not what appears in the pictures. None of the licenses above grants rights to third-party trademarks, copyrighted characters or a real person’s likeness, and several address this directly:
- Civitai’s content rules prohibit content that depicts or is based on the likeness of real people, living or dead, including public figures and private individuals. A celebrity LoRA found somewhere else is a red flag, not a loophole.
- The FLUX.1 [dev] license forbids uses that infringe third-party rights and names rights of publicity and “digital replica” rights.
- Open RAIL++-M forbids unlawful use, defamation and harassment.
- A LoRA trained to imitate a franchise can have permissive checkboxes and still produce images you can’t safely put in an ad: the checkboxes speak for the LoRA’s author, not for whoever owns the characters.
What is yours to use: your own brand, product, mascot or character, and photos you hold the rights to. If the content needs a recurring character, a LoRA trained on your own material avoids most of these questions; see character LoRA training.
Checklist before you launch
- List every model file in the template: checkpoint, LoRAs, VAE, upscalers, ControlNet, accelerator LoRAs. Each has its own license.
- For each Civitai file, open the exact version you use and note the License line.
- Open the permissions list. Selling outputs needs “Sell images they generate”; for contract or paid generation, also check “Run on services that generate for money”.
- Read the Addendum and the description: credit requirements, extra terms, a commercial licensing contact.
- Identify the base model and read its license at the source (Hugging Face model card and LICENSE file, the developer’s site), not a retelling.
- For FLUX.1 [dev] and anything built on it, decide before you start: a commercial license from BFL or a different base model.
- Check the content plan: no real people, no third-party brands or characters you have no rights to.
- Check the inputs: reference photos, product shots and image-to-video starting frames must be yours or licensed.
- Keep dated screenshots of model pages and copies of license texts with the project; where anything is unclear, get the author’s permission in writing or ask your lawyer.
How we check licenses
When a client sends a ComfyUI template, as a link to Civitai, OpenArt or GitHub or as a .json file, we go through the models and LoRAs in it by hand before anything runs, looking at the author’s permissions and the base model license of each. If a model is paid or gated, or its license doesn’t allow commercial use, we say so before the start, not after the batch is generated, and propose a way around it:
- a comparable checkpoint or LoRA with open permissions;
- a base model whose license allows commercial use, such as FLUX.1 [schnell] or SDXL for images;
- written permission from the author, since some authors list a contact for commercial licensing;
- a LoRA trained on your own character, mascot or product;
- if you don’t need the template itself, only the result, commercial video models via API, with their terms checked the same way.
We don’t give legal opinions. Our job is to make the license picture visible before money is spent; how you use the outputs is your decision.
When it’s easier to hand it to us
If you have a working template and need hundreds or thousands of outputs, we set it up on our GPUs, check the licenses of the models inside, run a trial batch for your approval, then generate at volume and review the results by hand: see ComfyUI workflow setup and bulk runs. To try a Civitai template yourself first, start with how to run a Civitai workflow. To have a specific template checked, use the request form.
