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:
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.
Agreed. There's also a lot of bad tinging/yellow saturation that very much reminds me of early gpt-image outputs on a lot of the non-cherry picked stuff I've been seeing on Twitter/Reddit.
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.
jjcm · · focus · HN ↗
<a href="https://html.non.io/qwen-comparison/" rel="nofollow">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
cloudking · · focus · HN ↗
BoorishBears · · focus · HN ↗
Even the artifacts are getting picked up.
vunderba · · focus · HN ↗
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.