[Feedback / Request] Revert Qwen-Image-2.1 to Apache 2.0 to Align with Qwen Family Licensing and Maximize Ecosystem Adoption
Dear Qwen Team and Alibaba Open Source Leadership,
First, congratulations on the release of Qwen-Image-2.1. The technical achievements—particularly the unified text-to-image capabilities, native 2K resolution, and efficient parameter footprint—are genuinely impressive, and the community response has been highly enthusiastic.
I am opening this discussion to share constructive feedback from the developer community regarding the model's licensing. While we appreciate the recent clarification that users retain rights to generated outputs, the underlying Qwen Research License (non-commercial only) still creates significant friction for developers, startups, and tool builders who want to integrate the model.
More importantly, this license represents a departure from Qwen's established open-source philosophy, and we believe there is a strategic opportunity to revert Qwen-Image-2.1 to an Apache 2.0 license.
Qwen's Proven Open-Source Track Record
Qwen has consistently demonstrated the power of permissive licensing—and it has worked exceptionally well.
Your entire ecosystem is built on Apache 2.0:
Latest models: Qwen 3.8 (just released) is Apache 2.0
Text models: Qwen3.5, Qwen3, Qwen2.5 series—all Apache 2.0
Vision-language models: Qwen2-VL, Qwen2.5-VL—all Apache 2.0
Previous image models: Earlier Qwen Image family releases were all Apache 2.0
These models have become the de facto standard in the open-source community. Developers trust, adopt, and build extensively on these models precisely because of the permissive license.
Qwen-Image-2.1 is the first to break this pattern.
This creates:
Confusion: Developers expect consistency across the Qwen family.
Friction: Tools and workflows built around previous Apache 2.0 Qwen Image models now face licensing barriers.
Lost momentum: The community was ready to build on 2.1 the exact same way they built on previous image models.
Your own ecosystem proves that Apache 2.0 works. It drives massive community adoption, integration into popular tools (Ollama, vLLM, ComfyUI), enterprise trust, and natural funneling to Qwen Cloud APIs. Qwen-Image-2.1 should return to this proven philosophy.
The Precedent: Google's Gemma Pivot
Google's earlier Gemma models faced community hesitation due to custom, restrictive licensing terms. Developers were concerned about redistribution limits, causing many to default to competitors. Recognizing that licensing friction was causing them to lose developer mindshare, Google pivoted. By releasing newer Gemma models under a permissive Apache 2.0 license, they eliminated this barrier. The result was immediate, massive community adoption, which ultimately drove more attention and usage to Google's broader ecosystem and APIs.
Alibaba is in an even stronger position—you've already proven Apache 2.0 works. Reverting to it is simply restoring what made Qwen successful in the first place.
Alibaba's Unique Advantage & The "Top-of-Funnel" Strategy
Alibaba is in a strong position to capitalize on this. You have a massive, integrated ecosystem including Alibaba Cloud and enterprise API services. Reverting Qwen-Image-2.1 to Apache 2.0 is not a loss of control; it is a powerful acquisition strategy:
Restore Community Trust: Developers who built on previous Apache 2.0 Qwen Image models will welcome the return to consistency.
Organic Ecosystem Growth: Developers will freely build UIs, fine-tunes, and integrations, creating organic buzz that makes Qwen the default choice.
Natural API Upsell: Developers who start by running the model locally will inevitably hit scale or infrastructure limits. When that happens, they naturally gravitate toward paid, managed Qwen Cloud API services.
An Apache 2.0 license changes this dynamic entirely. It removes the legal ambiguity and makes the global developer community work for you, not against you. We become your most effective R&D and marketing arm, driving adoption that ultimately monetizes through your cloud and API products.
Thank you for your continued innovation, transparency, and for listening to the community. We are excited to see what Qwen does next.
Qwen/Alibaba had been developing a reputation as being the go-to models for developers using open source licensed models. This is what built an ecosystem around them. They have a lot of downloads of this model due to the previous goodwill they had developed. But they are spending that goodwill, and the "next" company to come along and establish themselves as the go-to for open source will take their share.
"It takes 20 years to build a reputation and five minutes to ruin it."
Just an opinion, definitely open to being wrong, seems to me the license is deliberately muddy for a reason. Whats the reason? To cover their butts when someone uses it for something they shouldn't. This image model is more capable of combining and modifying images and that, precisely, makes it more useful for infringement loop holes and mimicking successful content creators. They aim to for only just enough similarity to make it convincing and get your click. An unfortunate reality of the digital landscape. Again, just an opinion. Especially when you consider all the lawsuits getting tossed around over AI.
I think this opinion is wrong. It took a lot of money to make the model, and if they do not make some kind of money off it, they are losing money. Also, since the models are trained on huge amounts of data, they are well able to make copyrighted characters. This has caused issues in the past.
I strongly support changing the Qwen-Image-2.1 license to Apache 2.0, or at least to a clearly permissive license that allows broader practical use.
The current research-only license is much more restrictive than many users may expect from a publicly downloadable model on Hugging Face. In practice, it excludes not only commercial SaaS products or paid image-generation services, but also many ordinary real-world use cases: artwork for websites, educational materials, calendars, books, promotional graphics, and content created by non-profit organizations.
That is especially unfortunate because Qwen-Image has built a strong reputation in the open model community, and previous Qwen image models were much easier to integrate into creative and technical workflows under Apache 2.0.
A permissive license would significantly increase adoption in ComfyUI workflows, local creative pipelines, open-source applications, research projects that later move into production, small businesses, artists, and non-profit organizations.
I hope the Qwen team will consider returning to Apache 2.0. It would make Qwen-Image-2.1 much more useful to the broader community and would likely increase adoption substantially.