Positives
• 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 to tackle this). Even though it's relatively trivial to set up background removal postprocessors, it's also neat to see it natively supported.
• It's fast using QwenImage2.1 convrot, a 1MP image took around ~5 seconds on an RTX4090.
Negatives
• The license (assuming you respect it) is far more restrictive. The original Qwen Image 1 was released under the standard Apache license; this one explicitly forbids commercial usage without obtaining a separate license. On the other hand, a lot of us didn't expect the Qwen team to ever release "weights-available" ever again.
Qwen-Image 1.0, released about a year ago, only scored 4/15 on my GenAI Showdown Benchmarks. Since that time, they've been upstaged by Krea 2 (6/15) and Ideogram4 (8/15). I'll post the new results once I have some more time to run them.
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 smaller, 7b vs 20b, it's multimodal (so you don't need a separate image-to-image model like you did with Qwen-Edit), more coherent, and significantly faster even when outputting at higher 2K resolutions. However, in my testing, I found that I had to play with dialing up the CFG depending on the complexity of the prompt.
I've also added a progress dropdown under Model Performance so you can see how cloud vs. local models have been trending since 2024. Spoiler: June of this year released some of the biggest bangers (Krea 2, Ideogram 4, and the kind of slept-on Boogu-Image 0.1).
Downsides:
- It was clearly trained on at least some level of synthetic training data, and it shows in some of the subpar outputs in terms of fidelity. Some of this you might be able to iron out with a refiner model downstream or a custom LoRA but time will tell.
- They've moved away from the permissive Apache license. Commercial usage is only allowed by request.
Comparisons:
https://genai-showdown.specr.net
If you just want to compare local models only:
I’ll have to think about this one. When I crafted the prompt, I wasn’t really thinking about the differences between a crucible and an anvil. It was more the visual of an archangel smelting halos for newly arrived heavenly beings.
Sorry, I'm not trying to nitpick. I'm just joining in because I'm interested in how the models dealt with the request.
Even though I prompted for a crucible in the prompt, I think the fact that the prompt also contained terms like “blacksmith” and “hammer,” caused it to lean towards anvils over crucibles in some of the pictures (which as you brought up makes more sense anyway).
Edit: This is wrong.
There is no single opinion, and clearly no single Chinese approach.
Also Chinese labs are in particular very careful about anything which be used to create pornographic content, which is highly illegal in the PRC.
MiniMax H3 is dominating AI porn right now, and they are Chinese.
The three benchmarks it failed on (D20, Flat Earth, and Banded Snake) are pretty difficult, so I'd be surprised if 2.5 manages to pass them, but I’ll add it for completeness’ sake later this week.
https://en.wikipedia.org/wiki/Qwen#List_of_models
Unfortunately, it looks like this model is using a much more restrictive license:
It's just freeware.
If AI labs get to ignore licenses, so do we.
Image data?
> GPL licensed instead and use it however the fuck I want
GPL is not a "use it however the fuck I want" license. Maybe you're thinking of the WTFPL?
If you were to generate outputs for commercial use, I think it would still violate this research license, but it's not like they are going to know, are they?
That said, I am disappointed that the model is not actually open-weights as I expected based on the headline.
It's not going to matter then, either. What are they going to do, sue me for copyright infringement?
Why even bother demanding such terms? Release the weights or don't release the weights, but this is silly. No one is going to pay them to run or host a 7B static image model.
It’s more of something to scare companies with legal teams. If you’re an individual or hobbyist doing a side project the risk is essentially zero.
People can continue to use closed SOTA models to generate outputs for commercial or malicious purposes.
What this research license achieves is that we cannot use this model in applications we publish.
How difficult would it be to use this model to create a second model without licensing issues?
And Krea 2 has a community license [2] that is fairly permissive - I think commercial usage is allowed under $1 million.
Boogu-Image scored 6/15 and Krea 2 scored 7/15 on my GenAI Showdown benchmark [3] - only Ideogram4 eclipses them in terms of local models, but its got a far more restrictive license and the JSON structured inputs can be a pain to work with.
[1] - https://github.com/Boogu-Project/Boogu-Image
[2] - https://www.krea.ai/krea-2-licensing
[3] - https://genai-showdown.specr.net/?models=fd,hd,kd,qi,f2d,zt,...
https://html.non.io/qwen-comparison/
The text rendering definitely is much, much better than anything else on the open weights market right now. Small text fidelity is quite good. It seems like the text encoder however gets a little bit overloaded with larger prompts - note the presence of hex codes in the design output, those were inputs from the expanded prompt.
I'll be trying a post-training run on this for web design, it has some serious potential.
[1] diffui.ai
That internal json backing helps significantly when you want to maintain consistent design system components/patterns across multiple pages. The aligned layout is it working as intended.
Even the artifacts are getting picked up.
A lot of people were putting ZiT as a refiner downstream in early Qwen-Image 1.0 workflows, so I'm wondering if we're going to see something similar with 2.1.
Really? Because basically everything in those screenshots is completely garbled. I didn't follow it super closely but I thought Ideogram or whatever was really good for this particular use, with actual clear text.
Recently, I've been helping a friend's wife with some basic vector images for her sewing hobby (she has what is essentially a CNC sewing machine) and have been super-impressed with FLUX.1-Kontext, which I've been running on my Macbook Pro with mflux. Its ability to (for example) take a photo of a human or an animal and return a line drawing which is recognisably them (rather than just a generic similarish image as I've experienced with other models) is excellent.
It's an older model now, but (AIUI) has the text-handling features baked in, and in my various testing is very reliable at giving me the outputs that I want, without the randomness I've experienced previously. It's big and relatively slow (~3 mins per 512x512 image edit on my M1 Max Mac) but excellent to work with. It's also very straightforward to set up, without the harness complexity of e.g. comfyui.
Through a combination of careful initial prompting (e.g. requesting line drawings, requesting limited colours ["posterization" works] if you need colour at all, requesting background removal) and then using the settings in VTracer to limit the colours used for the resulting SVG to the minimum needed, you can end up with an SVG that doesn't take too long to clean up in Inkscape. It usually takes a few minutes of deletions and path unions to produce something usable.
Diffusion models outputting SVGs would obviously be very interesting for this, but this is probably a small niche compared to everything that diffusion models can do with raster images, and I've not found anything usable online so far. LLMs (understandably, given their core technology) are still pretty bad at creating SVG art.
Prompt-adherence is really hit-or-miss—especially if one lacks the visual vocabulary. Likewise with coding, I find junior devs don't think to prompt re: respecting this-or-that interface, or refactoring to point-free style, etc.
So, as others have said, the artist knows better.
(I mean: outside direct or substantial use of Python, and running the Neural Network in the most efficient way.)
I tried stable-diffusion.cpp, following its compile guide here[0], and its Qwen Image-2.1 specific instructions here[1]. It works out of the box. I made a test pelican[2]. It took 3 minutes on a CPU.
[0] https://github.com/leejet/stable-diffusion.cpp/blob/master/d...
[1] https://github.com/leejet/stable-diffusion.cpp/blob/master/d...
total params memory size = 15645.19MB (VRAM 15645.19MB, RAM 0.00MB):
text_encoders 7669.77MB(VRAM),
diffusion_model 7331.05MB(VRAM),
vae 644.38MB(VRAM),
controlnet 0.00MB(N/A),
extensions 0.00MB(N/A)In fact, like it appears in the reports above, it is "7b" as in
> 7B parameters in its visual generation component
It seems they calibrated the size to fill a 16GB VRAM near the limit. RAM requirements will vary.
Edit x2: As usual I'm in a twisty maze of pip packages that don't work together, with obscure errors about missing modules, even though I followed the instructions on the page to the letter. I really wish people didn't use Python for this stuff. A simple C/C++ program would be so much better.
I think it will technically run on anything that has enough memory. I just tried it on a standard laptop (dual-channel DDR5), and it took about 3 minutes for a 512x512. If you'd want to run it at interactive speeds, you would want a GPU (one which fits this in VRAM).
> "I really wish people didn't use Python for this stuff. A simple C/C++ program would be so much better."
You mean besides stable-diffusion.cpp ?
Yes, thanks, I didn't know about that. Will try it.
it already has day-0 qwen image 2.1 support!
I have FLUX.2 klein and dev, Ideogram, LaDA-Image and SenseNova locally. Works great. Ive never touched a file.
The days of making container yamls myself is over. I read them but I dont edit anymore.
I think that's all Python (not a direct executable).
You could just do (see the "Quick Start") four `pip install` and have a dozen lines script to generate the image. But `llama.cpp` and similar do not require e.g. installing Torch (or PyTorch) - you can use `llama.cpp` on a non-specialized machine.
I don't think I have ever once run "pip install transformers" and had it work without three rounds of fiddling
Yep, that's (also) what I meant ;)
Lean, efficient... Also sensible and trouble-less.
> Currently, we support image, audio and video input.
Image outputs are supported, videos I'm not sure but I don't think that's an output, just a preview of the equirectangular example, so, same question here, what does this model outputs that isn't supported?
I understand that llama.cpp could only output text, last time I checked (I do not know how to find a good source for that though).
See https://github.com/ggml-org/llama.cpp/blob/master/src/llama-... , the
enum llm_arch {
At the risk of stating the obvious llama.cpp isn't just about LLaMa as https://github.com/ggml-org/llama.cpp/blob/master/src/llama-... someone else pointed out.
Back in Apr, ChatGPT Images 2.0 has some broken Chinese texts in its featured examples, and they later removed that from blog post. Is 2.5 better now?
EDIT: It still produces artifacts it's better but unusable for production work. In midvalues you will see a slight dot pattern.
Is this not simply some sort of watermark instead of an artifact?
Can you share the sources?
latents go from 16ch @ 8x compression to 64ch @ 16x, so roughly the same total latent budget but much more channel heavy. It’s also deeper/wider, and the old 2x2 transformer patching is gone.
On some images it still produces artifacts but can't say if it's the transformer or the VAE yet.
> You shall not use the Materials for any commercial purpose without obtaining a separate commercial license from us.
It probably will be much cheaper to use than other image models, but it seems that will be up to the whims of Qwen/Alibaba rather than just being the cost of putting it in a cloud provider.
Excited to see what the future holds for them!
I know a few friends of mine who are running models and are ignoring the licence.
Whether it is AGPL 3.0, or a completely restrictive license, it is going to get broken anyway and be used for commercial purposes.
I don't know anyone who looks at the licenses of the OSS software they are using.
In today’s world OSS is synonymous with "Free" and the AI model providers are proof of that with their training of code, datasets, etc.
So it begs the question, why should we abide by their licenses of their models?
Code on github, models on huggingface, nice intro text: "We are excited to open-source Qwen-Image-2.1 [...]".
meh...