# Qwen-Image

47 Hacker News commenters recommended Qwen-Image and 38 criticized it, November 2025 to November 2026. It was recommended in 10 of 13 months.

**Verdict:** Recommended for prompt following, text rendering and complex scenes among open-weights image models; the most common caveat is an unnatural smoothness and commercial use needs a separate license.

## Why people recommend it

- Strong prompt understanding and following (4 comments): "has significantly better prompt understanding but has an unnatural "smoothness" to its generated images" ([source](https://news.ycombinator.com/item?id=46175068)) "significantly better prompt understanding but has an unnatural "smoothness" to its generated images" ([source](https://news.ycombinator.com/item?id=46250763)) "The complex prompt following ability and editing is seriously impressive here." ([source](https://news.ycombinator.com/item?id=46957780))
- Open-weights release with diverse models (3 comments): "there's been a lot of positive movement in the open-weights space" ([source](https://news.ycombinator.com/item?id=48662879)) "Easily the most diverse set of models from all the Chinese labs." ([source](https://news.ycombinator.com/item?id=49775921)) "the Chinese models are easily the best or very close to the best" ([source](https://news.ycombinator.com/item?id=46111975))
- Strong text rendering, including CJK (2 comments): "The text rendering definitely is much, much better than anything else on the open weights market right now." ([source](https://news.ycombinator.com/item?id=49776897)) "A 7B diffusion model can now render CJK text better than Microsoft Windows." ([source](https://news.ycombinator.com/item?id=49775999))
- Handles complex layouts and scenes (2 comments): "Qwen-Image-2512 / Z-Image / Flux.2 absolutely crush SDXL if you're actually generating moderately complex scenes." ([source](https://news.ycombinator.com/item?id=48201311)) "effortlessly generating complex layouts such as newspapers, storyboards, and exam papers" ([source](https://news.ycombinator.com/item?id=48989834))

## What people warn about

- Unnatural smoothness in images (2 comments): "has significantly better prompt understanding but has an unnatural "smoothness" to its generated images" ([source](https://news.ycombinator.com/item?id=46175068)) "significantly better prompt understanding but has an unnatural "smoothness" to its generated images" ([source](https://news.ycombinator.com/item?id=46250763))
- License needs separate commercial deal (5 comments): "You shall not use the Materials for any commercial purpose without obtaining a separate commercial license from us." ([source](https://news.ycombinator.com/item?id=49776248)) "the license seems to cover any commercial usage at all, which would certainly include self-hosting for internal business purposes" ([source](https://news.ycombinator.com/item?id=49777556)) "now with the restriction on commercial use (which is against opensource definition) it is not even open-weights" ([source](https://news.ycombinator.com/item?id=49776620))
- Trails proprietary models like gpt-image (3 comments): "Everything is less good than gpt-2-image and I suspect that will be the case for awhile" ([source](https://news.ycombinator.com/item?id=49185018)) "It's a big step up over their prior model, but it's nowhere near gpt-image-2" ([source](https://news.ycombinator.com/item?id=49186965)) "From the images I've seen, Qwen-Image 3.0 is just a subpar equivalent to other proprietary models like gpt-image-2 and nb-pro." ([source](https://news.ycombinator.com/item?id=48993698))

## Alternatives hackers recommend

- [Nano Banana](https://hackerapproved.com/p/nano-banana-pro.md): 265 recommend
- [gpt-image-1](https://hackerapproved.com/p/gpt-image-1-5.md): 79 recommend
- [FLUX.2](https://hackerapproved.com/p/flux-2.md): 66 recommend
- [gpt-image-2](https://hackerapproved.com/p/gpt-image-2.md): 63 recommend
- [Midjourney](https://hackerapproved.com/p/midjourney.md): 37 recommend

## Cite as

Hacker Approved, "Qwen-Image", https://hackerapproved.com/p/qwen-image-2-1/ (counts from 1,795,640 Hacker News comments, November 2025 to November 2026).