Alibaba releases Qwen-Image-2.1, compact open-weight image model
The 7-billion-parameter model emphasizes efficiency and native transparency support but shifts to a more restrictive license than prior Qwen releases.
What to know
- Qwen-Image-2.1 cuts model size to 7 billion parameters—one of the smallest open-weight image generators available—while achieving competitive speed and introducing native transparency support.
- Text rendering substantially outperforms other open-weight models, making it attractive for UI/design workflows, but prompt-following reliability and visual artifacts remain limitations.
- The shift to a more restrictive license marks a departure from Qwen's prior open-licensing practice and raises questions about future openness of the project.
The dispute Whether the model's efficiency gains and text rendering capabilities outweigh its limitations in prompt following and visual artifacts for practical use cases. · positions read across 28 posts and comments
The model's efficiency and text rendering are genuine breakthroughs for open-weight image generation, especially for UI design.
-
“The text rendering definitely is much, much better than anything else on the open weights market right now.”
jjcm · Hacker News ↗
The restrictive license represents a concerning shift away from Qwen's prior open practices.
-
“Unfortunately, it looks like this model is using a much more restrictive license”
jfoster · Hacker News ↗
Despite improvements, the model still has significant production limitations including poor prompt adherence and remaining visual artifacts.
-
“My first impression is that it's not so good at following prompt directions…It instead gave me a broken 3D text on a white background.”
docheinestages · Hacker News ↗
Alibaba Qwen Team Developer
How it unfolded 7 developments, newest first · click a bar or a number to jump articlespostscomments
-
7
Technical analysis confirms efficiency gains and transparency support
Early testing by vunderba confirmed the model's significant size reduction (7b vs 20b parameters), native transparency support as a Qwen-exclusive feature, and fast inference speeds around 5 seconds for 1MP images.
“It's a heck of a lot smaller than Qwen-Image 1 (20b parameters) at only 7b, making it one of the smaller open-weight models available…It supports native transparency…”
— vunderba -
Well, the results are in, at least for text-to-image (the editing bench will come later).Qwen-Image 2.1 is definitely a pretty big leap over the last open-weight version, Qwen-Image 1.0, released back in August of last year and managed to score 7 out of 15 as opposed to its predecessor which scored 4 out of 15.Even though it's significantly…
2 more of the top 3 · 17 posts in this stretch
-
Meet Qwen-Image-2.1, the most balanced and cost-effective image generation model in the Qwen-Image series! Now open weights! 🎨 A unified model for both generation and editing, delivering top-tier quality in a lightweight package. Highlights: 👀 - Compact & exceptionally fast: A lightweight 7B ar...
-
So thoughtsPositives• It's a heck of a lot smaller than Qwen-Image 1 (20b parameters) at only 7b, making it one of the smaller open-weight models available (Z-Image Turbo is one of the few that is smaller at 6b) when compared to Ideogram, Krea2, Flux2, etc.• It supports native transparency (Qwen's team, as far as I know, is the only one attempting…
-
-
6
UI/design developers highlight superior text rendering as differentiator
Developer jjcm, who runs a prompt-to-UI design tool, tested Qwen-Image-2.1 and found its text rendering capabilities significantly better than other open-weight models, despite licensing concerns.
“The text rendering definitely is much, much better than anything else on the open weights market right now. Small text fidelity is quite good.”
— jjcm -
I run a prompt-to-ui design site that uses image models for the design process[1]. The text rendering especially makes this model deeply interesting to me, despite the license. Here are some tests using my harness comparing the outputs of gpt-image-2 and qwen 2.1:https://html.non.io/qwen-comparison/The text rendering definitely is much, much…
-
-
5
Early testers report struggles following complex prompt directions
User docheinestages reported that the model does not reliably follow detailed prompt instructions, requiring trial and error to achieve desired results rather than reliable directional control.
“My first impression is that it's not so good at following prompt directions…It instead gave me a broken 3D text on a white background.”
— docheinestages -
My first impression is that it's not so good at following prompt directions. I asked it to place a 3D text made of glass in a particular city. It instead gave me a broken 3D text on a white background. Maybe with different seeds it gets better, but it's more of a trial and error process than reliable results.
-
-
4
Users report VAE improvements but remaining artifacts limit production use
Commenter trentor noted that Qwen-Image-2.1 finally improved the VAE that had constrained prior models, but visual artifacts including dot patterns remain visible enough to make the model unsuitable for production work.
“They finally fixed their VAE. It really held back their models over the last 2 years.”
— trentor -
They finally fixed their VAE. It really held back their models over the last 2 years.EDIT: It still produces artifacts it's better but still unusable for production work. In midvalues you will see a slight dot pattern it's not as bad the older ones but still extremely visible.
1 more of the top 2 · 2 posts in this stretch
-
Its happy to see a new open image model from qwen. But the license is a let down. And it dosent even beat their closed qwen3 image wich is already a bit old.
-
-
3
Qwen-Image-2.1 uses more restrictive license than prior Qwen models
Users discovered that Qwen-Image-2.1 employs a more restrictive license compared to earlier Qwen models, which had used Apache and other more permissive licenses. This represents a licensing shift for the Qwen project.
“Unfortunately, it looks like this model is using a much more restrictive license…”
— jfoster -
A lot of the previous Qwen models seem to have used Apache licenses, among others:https://en.wikipedia.org/wiki/Qwen#List_of_modelsUnfortunately, it looks like this model is using a much more restrictive license:
1 more of the top 2 · 2 posts in this stretch
-
I am really grateful to the Chinese Labs for open sourcing their best models. If it was left to the Americans, we would be forced to pay obscene API fees to use them.
-
-
2
Users report strong local image generation performance relative to code generation
Hacker News commenter fishfasell noted that local text-to-image capabilities are currently outperforming local code generation in terms of speed and quality, contrasting general expectations about AI model capabilities.
“The capabilities of local LLM text-to-image is honestly pretty damn impressive…I can get an image in seconds locally with the quality being way higher than what I'd expect from a local model.”
— fishfasell -
Qwen-Image-2.1 is now supported in ComfyUI! Try it now and share your creations! 🖼️
2 more of the top 3 · 4 posts in this stretch
-
The capabilities of local LLM text-to-image is honestly pretty damn impressive. IMO, I think local image generation is currently ahead of local code generation. I can get an image in seconds locally with the quality being way higher than what I'd expect from a local model. However with coding it's much slower and much less impressive. I'm sure…
-
How do you use this model locally, similarly to using `llama-server -m <model>`?(I mean: outside direct or substantial use of Python, and running the Neural Network in the most efficient way.)
-
-
1
Alibaba releases Qwen-Image-2.1 image generation model
Qwen team announced Qwen-Image-2.1, a 7-billion-parameter open-weight image generation model designed as a more compact alternative to Qwen-Image 1 (20b parameters). The model supports native transparency and is positioned among the smallest open-weight image models available.
-
3 outlets first by TechNode, 5d ago · also RuntimeWire, Qwen · read ↗
-
2 outlets first by The Decoder, 5d ago · also Tom's Hardware · read ↗
-
Qwen-Image-2.1 is now supported in ComfyUI! Open weights. One 7B checkpoint that generates and edits. → Image generation at native 2K → Instruction editing from up to 10 reference images in a single pass → RGBA output, alpha included
-
What people are saying 11 voices from 2 sites · best of 28 · verbatim
- Sep 22
-
I get why they show mostly Chinese looking people, but why mostly women? I wanna see some handsome Chinese men as well!
- Sep 21
-
Now if we could get a Qwen-Image-Layered 2 also at 7B ... ^^(Or does this model has those capabilities natively ?)
-
I'm actually looking for a model that is able to create old school pixel art (like 90s style, games like Sierra and so on). But to this date, I haven't seen anything yet. It all reeks of "AI slop". Maybe this is a good thing, I don't know. But if anyone has any tips, I'd appreciate. It's for my own personal use.
-
Anyone get success editing videos, frame by frame, using image editing AI? Which model works well?In my experience, video models generate videos pretty well but are mid at editing. They actually regenerate the entire video along with the edit. So these models being non-deterministic tweak the rest of the video as well, the parts you hoped would be…
-
The minute you start looking at other non-frontier models, you understand why Dario, Altman, and Musk are coming together to say we need regulation.None are as good yet, but what everyone said is coming true - models are not moats. And these folks need an exit (even Msuk whose shares are still locked)
- Sep 20
-
Qwen just dropped a banger. Qwen-Image-2.1 is here, a 7B open-weight model for both image generation and editing. - Native RGBA support - Up to 10 reference images - Fast inference - Precise editing for portraits and products - Great for panoramas, infographics and virtual try-ons Open source ke...
-
Qwen Image 2.1 could be the best local model for image generation. Now open weights on HugginFace.
-
Impressive flex of image editing robustness from Qwen
-
Native transparency isn’t so hard to do by the way, I made an image AE (I don’t say VAE deliberately as none of these are VAEs, I don’t know why they keep being called that since the variational part is completely absent) that supported this about two years ago as a hobby project. I haven’t really been following the space recently, I’m surprised…
-
That benchmark might have some issues. You prompted the models to generate an image of striking a ring against a crucible. Then you (presumably, manually?) scored the images that depicted an anvil higher than the ones striking something resembling a crucible.
-
God I love the Qwen team. Easily the most diverse set of models from all the Chinese labs. Only Gemini/DeepMind comes close.