Is Commercial Use Allowed for the Illustrious-XL? License of SDXL Anime Models.
- Interpreting custom licenses is challenging
- SDXL models are likely commercially usable
- Over the long term, licenses trend stricter
Introduction
Hello, I'm Easygoing.
In this post, I'll be diving into the licensing of SDXL anime models.
For a comprehensive overview of licenses for image generation AI models, including next-generation ones like Flux.1 and Qwen-Image, check out my previous article.
The Conclusion First!
Since this is a long article, I'll start with the key takeaways right away.
gantt
title SDXL Anime Models
dateFormat YYYY-MM-DD
tickInterval 12month
axisFormat %Y
section Stability AI
Stable Diffusion XL: 2023-07-26, 2025-10-18
CreativeML Open RAIL++-M : milestone, 2023-07-26,
section Cagliostro Reserch Lab
Animagine-XL 3.0 :done, 2024-01-10, 2025-10-18
Animagine-XL 3.1 :done, 2024-03-17, 2025-10-18
Fair AI Public License 1.0-SD :done, milestone, 2024-01-10,
Animagine-XL 4.0 : 2025-02-13, 2025-10-18
section Onoma AI
Illustrious-XL v0.1 : done, 2024-09-25, 2025-10-18
Illustrious-XL v1.0 : 2025-02-11, 2025-10-18
Illustrious-XL v1.1 : 2025-02-18, 2025-10-18
Illustrious-XL v2.0 : crit, 2025-03-06, 2025-10-18
CreativeML Open RAIL-M: crit, milestone, 2025-03-06,
Gray: License prohibiting exclusive derivatives
Red: License granting rights to modify derivatives and their outputs
- Outputs from SDXL are likely usable for commercial purposes
- The Illustrious-XL v2.0 model allows the rights holder (Onoma AI) to modify derivatives and their outputs
- Custom licenses are trending toward stricter terms
With that said, let's walk through the licensing of SDXL anime models step by step.
Local Image Generation Started with Stable Diffusion 1
Local image generation kicked off with the release of Stable Diffusion 1 in August 2022.
gantt
title Image Generation AI
dateFormat YYYY-MM-DD
tickInterval 12month
axisFormat %Y
section Open AI
Dall-E : 2021-01-05, 2025-10-18
section Midjourney
Midjourney : 2022-02-01, 2025-10-18
section Google
Imagen : 2022-05-23, 2025-10-18
section Stability AI
Stable Diffusion 1: crit, 2022-08-22, 2025-10-18
At the time Stable Diffusion 1 launched, models like OpenAI's DALL-E and Google's Imagen had already been developed, but they were limited to research use and not available to the general public.
In this landscape, up-and-coming AI startup Stability AI made waves worldwide by releasing the cutting-edge Stable Diffusion 1 image generation model in August 2022 under a license that permits copying, modification, and commercial use.
CreativeML Open RAIL-M License
Stable Diffusion 1 came with the CreativeML Open RAIL-M license.
gantt
title Stable Diffusion Series
dateFormat YYYY-MM-DD
tickInterval 12month
axisFormat %Y
section Model
Stable Diffusion 1: crit, 2022-08-22, 2025-10-18
Stable Diffusion 2: 2022-11-24, 2025-10-18
Stable Diffusion XL: 2023-07-26, 2025-10-18
section Licence
CreativeML Open RAIL-M: crit, milestone, 2022-08-22,
CreativeML Open RAIL++-M: milestone, 2022-11-24,
CreativeML Open RAIL-M License
- Free use of the model and disclaimer of liability for the copyright holder (Stability AI)
- Permits copying, modification, and commercial use
- Allows strictening derivative licenses but not loosening them
The CreativeML Open RAIL-M license was Stability AI's custom license, crafted specifically for Stable Diffusion 1's release.
- Standard licenses → User-focused
- Custom licenses → Tailored by the copyright holder to their goals
Under the CreativeML Open RAIL-M license, the model can be used freely at the user's own risk, while establishing Stability AI's liability exemption.
The Stable Diffusion 1 model allows copying, modification, and commercial use for reasonable purposes.
For derivatives, you can stricten the license, enabling the creation and launch of paid services with improved models.
Example: Novel AI, which enhanced Stable Diffusion to offer an anime illustration generation service
On the flip side, the CreativeML Open RAIL-M license does not allow loosening the license for derivatives (since a looser license on copies would undermine the original's purpose).
Thus, derivatives of Stable Diffusion 1 inherit the CreativeML Open RAIL-M license.
Stability AI Held Rights to Modify Derivatives
Once Stable Diffusion 1 and its CreativeML Open RAIL-M license were released, communities on Discord and YouTube buzzed with discussions.
A major point of contention was that Stability AI held rights to modify derivatives and their outputs.
CreativeML Open RAIL-M (SD1)
7. Updates and Runtime Restrictions.
To the maximum extent permitted by law, Licensor reserves the right to restrict (remotely or otherwise) usage of the Model in violation of this License, update the Model through electronic means, or modify the Output of the Model based on updates. You shall undertake reasonable efforts to use the latest version of the Model.
While Stable Diffusion 1 promoted free development, Stability AI's ability to manage derivatives led to criticism that true free development was hindered.
CreativeML Open RAIL++-M License Removes Remote Modification Rights
In response to the backlash, Stability AI released a revised CreativeML Open RAIL++-M license with the problematic clause removed.
CreativeML Open RAIL++-M (SD2, SDXL)
Section IV: OTHER PROVISIONS
Updates and Runtime Restrictions. To the maximum extent permitted by law, Licensor reserves the right to restrict (remotely or otherwise) usage of the Model in violation of this License.
The CreativeML Open RAIL++-M license eliminated Stability AI's remote modification rights for models and outputs, enabling even freer development.
This license was adopted for the newly released Stable Diffusion 2 and Stable Diffusion XL, making SDXL the most freely developable image generation AI model from day one.
Commercial Success of Novel AI and Leonardo AI
From the SDXL model sprang numerous derivatives, leading to commercial hits like Novel AI and Leonardo AI.
These derivatives outperformed the original SDXL, so Stability AI—which provided the base model—didn't achieve commercial success itself.
However, Stability AI never claimed rights infringement against competitors, honoring the CreativeML Open RAIL++-M license.
Just 2.5 years since its debut, SDXL's reputation as commercially usable is solidifying, thanks to Stability AI's stance.
The Arrival of Animagine-XL 3.0!
In January 2024, China's Cagliostro Research Lab released Animagine-XL 3.0, fine-tuned on the original SDXL with 1.3 million anime illustrations.
Animagine-XL 3.0 boasted unprecedented illustration quality, becoming the driving force behind SDXL anime model adoption.
Fair AI Public License 1.0-SD Prohibits Exclusivity
Animagine-XL 3.0 uses the Fair AI Public License 1.0-SD, which adds custom restrictions to CreativeML Open RAIL++-M.
Restrictions in Fair AI Public License 1.0-SD
- If modified, disclose the modifications
- For online services, make the model downloadable by users
Notices
If you modify this software and allow users to interact with it through a computer network, you must ensure they have a reasonable way to receive the corresponding source code from you, whether that is via a download link or a prominent written offer. As a special case, if you are only allowing users to interact with a derived model, then you may choose to provide a download link or written offer only for the derived model.
The Fair AI Public License 1.0-SD goes a step further than CreativeML Open RAIL++-M by requiring model publication.
This is a strong counter to exclusive paid online services like Novel AI, emphasizing Animagine-XL as an open model prohibiting exclusivity.
The Debut of Illustrious-XL!
In September 2024, Korea's Onoma AI developed Illustrious-XL by fine-tuning Kohaku XL-Beta Revision 5 on 7.5 million images.
gantt
title SDXL Anime Models
dateFormat YYYY-MM-DD
tickInterval 12month
axisFormat %Y
section Stability AI
Stable Diffusion XL: 2023-07-26, 2025-10-18
CreativeML Open RAIL++-M : milestone, 2023-07-26,
section Cagliostro Reserch Lab
Animagine-XL 3.0 :done, 2024-01-10, 2025-10-18
Animagine-XL 3.1 :done, 2024-03-17, 2025-10-18
Fair AI Public License 1.0-SD :done, milestone, 2024-01-10,
Animagine-XL 4.0 : 2025-02-13, 2025-10-18
section Onoma AI
Illustrious-XL v0.1 : done, 2024-09-25, 2025-10-18
Illustrious-XL v1.0 : 2025-02-11, 2025-10-18
Illustrious-XL v1.1 : 2025-02-18, 2025-10-18
Illustrious-XL v2.0 : crit, 2025-03-06, 2025-10-18
CreativeML Open RAIL-M: crit, milestone, 2025-03-06,
Illustrious-XL v0.1 was announced for paid release at $5/month starting September 15, 2024—but on September 12, it leaked via closed beta testers.
Quoted from official paper
Onoma AI filed a DMCA takedown, removing the leak by September 25. But copy sites proliferated immediately, prompting Onoma AI to release the model for free on Hugging Face on September 21, 2025.
Illustrious-XL v0.1 adopted the same Fair AI Public License 1.0-SD as Animagine-XL, thrilling the community with another high-performance model's open release.
Illustrious-XL 1.0 Goes Paid Again
Illustrious-XL v0.1's quality won widespread support and spawned many derivatives.
Anticipation built for Illustrious-XL v1.0, but Onoma AI announced paid distribution to recoup development costs.
Illustrious-XL 1.0 Paid Service Details
- Closed paid distribution ($10)
- Set a free-release goal and solicit donations on the official site
- Paid online image generation service
Does Illustrious-XL v1.0 Violate the License?!
flowchart TB
O1(SDXL_Base<br>CreativeML Open RAIL++-M)
A1(Animagine-XL 3.0<br>2024.1.10<br>Fair AI Public License 1.0-SD)
A2(Animagine-XL 3.1<br>2024.3.17<br>Fair AI Public License 1.0-SD)
B1(Animagine-XL 4.0<br>2025.2.13<br>CreativeML Open RAIL++-M)
C1(Kohaku XL-Beta Revision 5<br>2023.11.2<br>CreativeML Open RAIL++-M)
C2(Illustrious-XL v0.1<br>2024.9.25<br>Fair AI Public License 1.0-SD)
C3(Illustrious-XL v1.0<br>2024.2.11<br>CreativeML Open RAIL++-M)
C4(Illustrious-XL v1.1<br>2024.2.18<br>CreativeML Open RAIL++-M)
D1(Illustrious-XL v2.0<br>2024.3.6<br>CreativeML Open RAIL-M)
O1--->A1
O1------>B1
O1-->C1
O1-------->D1
A1-->A2
C1---->C2
C2-->C3
C3-->C4
Illustrious-XL v1.0's license has two major issues:
- Looser license than previous Illustrious-XL v0.1
- Violates the exclusivity-prohibiting Fair AI Public License 1.0-SD
First, v1.0 uses the looser CreativeML Open RAIL++-M, violating the rule against loosening derivative licenses.
Second, Onoma AI's paid online service for v1.0 directly matches the scenario in Fair AI Public License 1.0-SD, requiring downloadable model access for users.
Onoma AI claimed:
- Fair AI Public License 1.0-SD encourages open-sourcing derivatives for contributors but doesn't impose strict "non-commercial" rules
- They'd release for free after recouping costs
But this argument wasn't widely accepted.
For reference, Novel AI also freely releases past models for non-commercial use after cost recovery.
Is the License Violation a Problem?
Illustrious-XL v1.0 likely violates Fair AI Public License 1.0-SD—but is that an issue?
When a violation is suspected, the rights holder ultimately decides.
Here, the rights holder for v0.1's Fair AI Public License 1.0-SD is Onoma AI itself, so it won't face enforcement.
(If the base Kohaku XL-Beta Revision 5 had Fair AI Public License 1.0-SD, the judgment might differ.)
However, Onoma AI's self-interpretation undermined trust in Fair AI Public License 1.0-SD.
Thus, the February 2025 Animagine-XL 4.0dropped it.
Illustrious-XL v1.0 Leaks Again!
Illustrious-XL v1.0's paid release sparked massive debate.
On launch day for closed paid distribution, a paying user leaked it again.
Onoma AI demanded takedown, but the leaker refused.
Regardless of developer-leaker ethics, v1.0's CreativeML Open RAIL++-M permits copying and publishing the model.
Official Free Release Calms the Storm
Donations for Illustrious-XL v1.0 fell short of the goal, but the official free release in February 2025 ended the uproar.
The follow-up Illustrious-XL v1.1 also missed its donation goal but was released for free.
Illustrious-XL v2.0 Surpasses Traditional SDXL Expressiveness
In April 2025 (at this point, the donation rationale feels unclear), the latest Illustrious-XL v2.0 was released.
Illustrious-XL v2.0 generates high-resolution illustrations exceeding standard SDXL, with detail expressiveness rivaling or surpassing next-gen models.
Illustrious-XL v2.0 Lets Onoma AI Modify Derivatives!
However, v2.0 reverts to Stable Diffusion 1-era CreativeML Open RAIL-M.
Unlike SDXL, it grants the rights holder (Onoma AI) rights to modify derivatives and outputs, as explained earlier.
Since v2.0 is a from-scratch retrain (not an improvement on prior models), a stricter license makes sense.
flowchart TB
O1(SDXL_Base<br>CreativeML Open RAIL++-M)
A1(Animagine-XL 3.0<br>2024.1.10<br>Fair AI Public License 1.0-SD)
A2(Animagine-XL 3.1<br>2024.3.17<br>Fair AI Public License 1.0-SD)
B1(Animagine-XL 4.0<br>2025.2.13<br>CreativeML Open RAIL++-M)
C1(Kohaku XL-Beta Revision 5<br>2023.11.2<br>CreativeML Open RAIL++-M)
C2(Illustrious-XL v0.1<br>2024.9.25<br>Fair AI Public License 1.0-SD)
C3(Illustrious-XL v1.0<br>2024.2.11<br>CreativeML Open RAIL++-M)
C4(Illustrious-XL v1.1<br>2024.2.18<br>CreativeML Open RAIL++-M)
D1(Illustrious-XL v2.0<br>2024.3.6<br>CreativeML Open RAIL-M)
O1--->A1
O1------>B1
O1-->C1
O1-------->D1
A1-->A2
C1---->C2
C2-->C3
C3-->C4
Low Chance of Actual Modification, But...
Remotely rewriting models and outputs is technically challenging, so with current tech, actual modification is highly unlikely.
CreativeML Open RAIL-M poses no issues for personal illustration generation and enjoyment.
However, if commissioned to create and deliver commercial illustrations using a v2.0 derivative, residual modification rights on deliverables could be a concern.
Custom Licenses Serve Specific Purposes
Custom licenses are set to align with the rights holder's goals.
As seen with Stability AI, Cagliostro Research Lab, and Onoma AI, interpretations vary by holder.
Given a custom license, anticipating later strictening and paid models as adoption grows is business as usual.
After 2.5 years for SDXL and about 1 year for Illustrious-XL, commercial usability is gaining traction—but a single court ruling could overturn it.
For commercial illustrations under strict licenses, I recommend models with open standard licenses for safety.
Models with Open Standard Licenses
- AuraFlow, Flux.1 [schnell], HiDream, Qwen-Image
Summary: Custom Licenses Are Tricky
- Interpreting custom licenses is challenging
- SDXL models are likely commercially usable
- Over the long term, licenses trend stricter
Illustrious-XL's paid releases sparked heated community debate with mixed opinions.
Still, no one disputes Onoma AI's generosity in freely releasing costly trained models. Illustrious-XL and its derivatives are undeniably among the community's most supported anime models today.
New tech like image generation AI always evolves in gray areas.
Why not explore your own interpretations of complex licenses when using AI models?
Thanks for reading to the end!