List of Licenses for Image Generation AI Models: Can This Model Be Used Commercially?

Introduction

Hello, I’m Easygoing!
In this post, we’ll dive into the essential topic of licenses that you need to understand when using image generation AI.

License = Restrictions on Copyright Holder’s Rights

A license corresponds to restrictions on the rights of the copyright holder.

Computer programs, including image generation AI, are considered creative works and are subject to copyright law.

A license is a statement by the copyright holder that using the work for specific purposes does not constitute copyright infringement.

In other words, it’s a way for the copyright holder to limit their own rights.

AI and Copyright

Open Licenses Drive Adoption!

While licenses restrict the rights of the copyright holder, they also offer benefits to those who set them.

Namely, models with open licenses tend to gain widespread adoption.

Models with open licenses attract more users, increasing their market share. Additionally, community-driven development leads to the creation of more advanced models.

Standard Licenses vs. Custom Licenses

Licenses can be divided into standard licenses, which have been widely recognized for a long time, and custom licenses, which are specifically tailored for individual models.

• Standard licenses → User-friendly
• Custom licenses → Set by the copyright holder for specific purposes

Let’s explore some specific examples of each.

Standard Licenses Are Easy to Understand!

Standard licenses are straightforward and user-friendly.

Standard licenses are globally recognized, with numerous precedents for their application. They are highly reliable and unlikely to change in the future.

Common Standard Licenses

The following are some of the most common standard licenses used for image generation AI models:

gantt
    title Standard License
    dateFormat YYYY
    axisFormat %Y

    section Commercial License
        MIT : a1, 1984, 2026
        Apache-2.0 : a2, 2004, 2026
        WTFPL : a3, 2000, 2026
        CC0   : a4, 2009, 2026

    section Non Commercial License
        CC-BY-NC-SA-4.0 : done, b1, 2013, 2026

Open Standard Licenses

The four licenses listed below, while differing in specifics, generally permit free use of the model, including for commercial purposes.

MIT（Massachusetts Institute of Technology）

• Allows unrestricted use, copying, modification, distribution, and sale of the software.
• The software is provided “as is,” with no liability for damages borne by the provider.
• Requires including the full text of the MIT License when distributing.

Apache 2.0 （Apache License, Version 2.0）

• Provides a perpetual, worldwide, non-exclusive copyright license.
• Permits copying, modification, distribution, and sublicensing.
• Compared to MIT, it includes explicit patent grants and obligations to indicate derivative works.

WTFPL（Do What The Fuck You Want To Public License）

• Despite its humorous name, it is often deemed legally valid.
• Allows redistribution and modification, with nearly all copyright holder rights waived.
• Lacks a warranty disclaimer (rarely used by corporations).

CC0（Creative Commons Zero）

• Completely waives copyright, or permits maximum free use in countries where full waiver isn’t possible.
• Users can copy, modify, distribute, and use commercially without needing permission.
• No warranties are provided for the work.

Models with these four licenses can be used commercially with confidence.

Licenses Restricting Commercial Use

On the other hand, some standard licenses explicitly restrict commercial use, with CC-BY-NC-SA-4.0 being a prominent example.

CC-BY-NC-SA-4.0（Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International）

• Permits free use of the work but prohibits commercial use.
• Requires crediting the original author.
• Modifications are allowed, but the same license must be inherited.

The CC-BY-NC-SA-4.0 license is used by some AI upscaling models.

Since this license explicitly prohibits commercial use, models with this license should not be used for commercial purposes.

Custom Licenses Are Set by Copyright Holders

While standard licenses are widely accepted, custom licenses are created by copyright holders for specific purposes.

Custom licenses allow more flexible conditions, making them ideal for cases where copyright holders want to promote open use while retaining certain rights.

However, custom licenses are often subject to change and have fewer legal precedents, making them less reliable than standard licenses.

Custom Licenses Used in Image Generation AI

gantt
    title Custom License
    dateFormat YYYY-MM-DD
    tickInterval 12month
    axisFormat %Y

    section Stable Diffusion 1
        CreativeML Open RAIL-M : a1, 2022-08-22, 2025-09-05

    section Stable Diffusion XL
        CreativeML Open RAIL++-M : b1, 2022-11-24, 2025-09-05
        Fair AI Public License 1.0-SD :  b2, 2023-11-23, 2025-09-05

    section Stable Diffusion 3
        Stability AI Community License :  c1, 2024-04-14, 2025-09-05
        
    section Flux.1 [dev]
        FLUX.1 [dev] Non-Commercial License : done, d1, 2024-08-01, 2025-09-05

CreativeML Open RAIL-M

• Permits modification, redistribution, and commercial use.
• Prohibits use for inappropriate purposes.
• Ownership and responsibility for outputs lie with the user.

The CreativeML Open RAIL-M license was created by Stability AI alongside the release of Stable Diffusion 1, emphasizing responsible AI use and open development.

CreativeML Open RAIL++-M

• A revised version of CreativeML Open RAIL-M.
• Removes the rights holder’s (Stability AI) ability to remotely update or modify outputs.
• Promotes greater community-driven use.

The CreativeML Open RAIL++-M license was revised based on community feedback, enabling freer development by eliminating Stability AI’s rights to remotely update or modify models.

Fair AI Public License 1.0-SD

• Largely compatible with CreativeML Open RAIL++-M.
• Requires public disclosure of source code for modified parts during modification or redistribution.
• Prohibits exclusive control of derivative models.

The Fair AI Public License 1.0-SD was introduced in November 2023 for the Animagine-XL 3.0 model. It advances openness by requiring disclosure of modified parts and ensuring models remain publicly accessible.

Stability AI Community License

• Applies to Stable Diffusion 3.
• Free for individuals and small businesses.
• Users with annual revenue exceeding $1 million must purchase a commercial license.

The Stability AI Community License, created for Stable Diffusion 3, grants free use to individuals and small businesses but requires large businesses to purchase a commercial license.

FLUX.1 [dev] Non-Commercial License

• A non-commercial license intended for research purposes.
• Allows modification and redistribution but prohibits using outputs to train competing models.
• Commercial use of outputs is ambiguously described (generally prohibited).

The FLUX.1 [dev] Non-Commercial License applies to Black Forest Labs’ Flux.1 [dev] series. While it permits free use and modification, it prohibits using outputs to train other models.

Commercial Use of Flux.1 [dev] Is Ambiguous

The FLUX.1 [dev] Non-Commercial License is highly ambiguous regarding commercial use:

1. The license is named “Non-Commercial License.”
2. Version 1.0 initially included a clause permitting commercial use of outputs.
3. This clause was removed in the version 1.1 revision on June 27, 2025.
4. The following day, version 1.1.1 reinstated the previous wording.

Licenses like FLUX.1 [dev] Non-Commercial License, which undergo frequent changes to critical sections, tend to lack reliability.

While interpretations of commercial use for Flux.1 [dev] vary, I believe the license is intended to prohibit commercial use.

Illustrious-XL Series Has Complex Licensing

The Illustrious-XL series, a popular anime model based on SDXL released by South Korea’s Onoma.AI, has different licenses for each version. Here’s a summary of the key points:

• Initial version (ver 0): Requires disclosing modified parts and making the model publicly available.
• Latest version (ver 2.0): The rights holder (Onoma.AI) retains rights to modify derivative models and outputs.

The Illustrious-XL series includes numerous derivative models, some of which may combine multiple versions. When using the Illustrious-XL series or its derivatives, carefully review the applicable licenses beforehand.

Are License Revisions Applied Retroactively?

While custom licenses may be revised, it’s generally understood that new licenses do not apply retroactively to models already released under older licenses.

However, there are few legal precedents for custom licenses, and some licenses or terms of use may include clauses suggesting future revisions.

Keep in mind that custom licenses are set at the discretion of the copyright holder, allowing significant flexibility in changes and scope of application.

Models Without Licenses

For locally used models, licenses like those mentioned above are typically applied to encourage adoption.

However, for services provided online by large corporations, where sufficient adoption is expected without a license, licenses are often not applied.

Examples of Models Without Licenses

• Open AI: GPT-image-1
• Google: Imagen, Gemini-Flash-Image (nano-banana)
• X.ai: Aurora

Since these models lack licenses, all copyright remains with the rights holder.

When using these models, there are no license-based guarantees for users, so it’s critical to thoroughly review the terms of use to understand permissible uses.

Online Services May Reuse Data

Many online image generation services specify in their terms that input and output data may be reused for service promotion or AI training.

When creating illustrations for clients using online services, it’s preferable to use services or plans that explicitly prohibit data reuse to avoid unintended data leaks.

Don’t Disclose the Model Used

When delivering commercial illustrations to clients, it’s generally safer not to disclose the model used.

This is because future changes to the model’s terms could render commercial use impossible, potentially placing responsibility on the client to address any issues.

When generating AI illustrations, keeping records of models and prompts is essential to avoid future issues, but unless specifically requested by the client, such information should be withheld.

Summary: Standard Licenses Are Reliable!

• Standard licenses are trustworthy.
• Custom licenses are subject to change.
• Don’t disclose the model used.

Image generation AI models come with various licenses. Models with open standard licenses, such as AuraFlow, Flux.1 [schnell], HiDream, and Qwen-Image, allow their outputs to be used commercially with confidence.

On the other hand, models with custom licenses can be difficult to interpret, and no definitive answers exist.

For models with custom licenses, consider why a standard license wasn’t used and what the rights holder aims to protect. This may provide some clarity.

Thank you for reading to the end!