SD3 License Is Insane!
The release of Stable Diffusion 3, the latest text-to-image AI generator model from Stability AI, has sparked controversy within the AI community due to concerns over its usage license. This blog post explores the key issues and criticisms surrounding the license’s terms and conditions, and their potential impact on the AI community.
SD3 Medium was released on June 12, 2024. It is a 2 billion parameter text-to-image model designed to excel in areas where previous models struggled, particularly in photorealism, typography, and fine-tuning capabilities.
The main differences between SD3 Medium and previous models include:
- Photorealism: SD3 Medium overcomes common artifacts in hands and faces, delivering high-quality images without complex workflows.
- Typography: SD3 Medium provides powerful typography results that surpass the latest large models, making it ideal for detailed text rendering.
- Performance: SD3 Medium is optimized for both consumer systems and enterprise workloads, with a low VRAM footprint that allows it to run efficiently on consumer GPUs.
- Fine-Tuning: SD3 Medium can absorb fine details from small datasets, making it perfect for customization and creativity.
- Architecture: SD3 Medium employs new noise samplers that emphasize perceptually relevant scales, leading to superior performance over traditional diffusion methods. It also supports varying resolutions and aspect ratios through adaptable positional encodings.
- Commercial Use: SD3 Medium is available for non-commercial use only, with a self-hosting license required for commercial use. This is a significant change from previous models, which were often available for free.
These advancements make SD3 Medium a significant improvement over previous models, particularly in terms of photorealism, typography, and fine-tuning capabilities.
Overview of Stable Diffusion 3 License
The Creator’s License
Stability AI has introduced a “Creator’s License” for the Stable Diffusion 3 model, which applies to creators and developers with less than $1 million in annual revenue, less than $1 million in institutional funding, and fewer than 1 million monthly active users.
License Fees and Image Generation Limits
The Creator’s License comes with a $20 per month fee, even for those running the models locally on their computers. Surprisingly, the license reportedly limits image generation to a mere 6,000 images per month, which many argue is an unreasonably low limit.
Excuse me, but what?!?
First of all, I’ve tried the SD3 medium, and let me tell you, it’s not easy to get a usable result on the first try.
So, if you are a company that want to provide SD3 for your users and have for example 500k active users, you will have to limit them to a TOTAL of 6,000 image per month. So, basically, you cannot offer it without frustrating everybody.
I’ve talked to many active influencers in the AI niche and also with online graphic design app stakeholders that use integrated Stable Diffusion models that they are also staying away from SD3 for now at least.
Derivative Works and Copyright Issues
The license defines “derivative works” as any modifications to the core model, including models created based on or derived from the core model or its output. This raises concerns about copyright implications and the potential impact on the broader AI community.
Major Concerns and Criticisms
One of the most important communities of AI creators, CivitAI has recently banned Stable Diffusion 3 and derivates from its website: https://civitai.com/articles/5732
This is a clear sign that until Stability AI clarifies its intentions with this model, the open source community is taking a big step back.
Unreasonable Image Generation Limits
Critics strongly argue that the 6,000 images per month limit is absurdly low, especially given the current state of Stable Diffusion 3, which is described as an early development model with numerous mistakes and inferior performance compared to previous models.
Destruction of Derivative Works upon License Termination
One of the most contentious aspects of the license is the requirement to destroy all derivative works, including trained models and their outputs, upon termination or expiration of the license. This means that users and their customers must delete all models they have trained based on Stable Diffusion 3, even if the termination is due to financial constraints.
I’m talking about model fine tunings. These have been essential for establishing SD1.5 and SDXL models as the norm for open source AI image generations. Models like Dream Shaper, Juggernaut XL, Turbo & Lightning and hundreds of other fine tunings of the base models have been created by the communities and posted on Hugging Face or CivitAI.
I think it’s fair to say that community created models have been the main driver of Stable Diffusion’s popularity as an AI image generator.
The SD3 new license makes impossible to follow the same path. So, if you create an SD3 fine tuned checkpoint, you are obligated delete all the models and your users models if you stop paying the license fee.
How do you spell ABSURD?
Liability for Customer and User Actions
Concerns have been raised about the clause that holds users liable for any actions or omissions by their customers or users in connection with their use of the software product, suggesting that users should only be responsible for their own actions.
Potential for Future Fee Increases
There are worries about the potential for Stability AI to increase fees in the future, effectively holding users and their customers hostage if they have built their businesses around Stable Diffusion 3 and its derivative works.
Supporting the AI Community and Preserving Open Source Values
Here are my thoughts on what Stability AI should consider.
Honoring Small Artists and Designers
In my opinion, Stability AI should recognize and honor the contributions of small artists and designers, who have invested significant time and effort into improving the models and giving them value. Stability AI should exempt or provide more favorable terms for these individuals.
Setting Higher Revenue Thresholds for Licensing
Instead of targeting creators and developers with less than $1 million in annual revenue, it has been proposed that Stability AI should set higher revenue thresholds, such as $500,000 or more, for licensing fees. This would allow smaller entities to continue contributing to the AI community without facing financial barriers.
Rethinking the License Terms
Overall, there is a call for Stability AI to rethink and rephrase its license agreement, ensuring that it aligns with the values and principles of the open source community, which has played a crucial role in the development of these AI models.
Can We Do Anything? Advocating for Change
The concerns raised by the AI community regarding Stable Diffusion 3’s usage license are significant, and addressing them will require a concerted effort. Here are some strategies the community could employ to advocate for changes to the license terms:
Organize a Unified Voice
One of the most effective ways to gain Stability AI’s attention is through a coordinated effort that represents a substantial portion of the AI community.
This could take the form of:
- Petitions and Open Letters — A well-crafted petition or open letter, signed by influential members, developers, researchers, and artists, would carry substantial weight and demonstrate the widespread concerns within the community.
- Direct Outreach — Respected community members could attempt to engage directly with Stability AI’s leadership or legal team. Presenting well-reasoned arguments and highlighting the potential negative impacts on the community’s growth and innovation could persuade the company to reconsider its stance.
You can reachout to Stability AI via:
Leverage Public Pressure
The community could also leverage social media platforms and public forums to raise awareness about the issues and garner broader support.
- Social Media Campaigns - Hashtag campaigns, blog posts, and discussions on popular forums could generate public pressure and media attention, potentially influencing Stability AI’s decision-making process
- Media Outreach — Reaching out to media outlets and journalists covering the AI and technology sectors could help amplify the community’s concerns and increase public scrutiny on Stability AI’s license terms.
Propose Alternatives
Instead of solely criticizing the current license, the community could propose alternative license models that better align with open source principles and the needs of different user groups:
- Open Source-Friendly Licenses — Presenting well-thought-out alternatives, such as licenses used by other successful open source projects, could make it easier for Stability AI to consider modifications.
- Tiered Licensing Structure — The community could suggest a tiered licensing structure that accommodates different user groups, such as hobbyists, small businesses, and large enterprises, with varying levels of restrictions and fees.
Loss of Community Engagement
Stability AI may be more inclined to make changes if they understand the potential long-term consequences of losing community engagement and the collaborative ecosystem that has played a crucial role in the development of these AI models.
Seek Support from Influential Organizations
Reaching out to influential organizations within the AI and open source communities, such as the Open Source Initiative (OSI) or the Electronic Frontier Foundation (EFF), could lend additional weight and credibility to the concerns raised by the community.
While a peaceful and persistent approach is recommended, the community could explore legal avenues as a last resort if Stability AI remains unresponsive to the concerns raised and the proposed alternatives.
Conclusion: Call for Transparency and Community Engagement
In conclusion, the concerns raised highlight the need for Stability AI to engage transparently with the AI community, address the issues raised, and reconsider the terms of its usage license. Preserving open source values, supporting small contributors, and fostering a collaborative ecosystem that benefits all stakeholders are paramount. By addressing these issues, Stability AI can build trust and maintain a thriving AI community committed to advancing this cutting-edge technology.
I’m curious what you think. Drop me a comment below with your thoughts on this.
