[mattnewton]

Krea 2 Large (on the website and api) was trained with the FLUX 2 VAE, if you want to test it out and push realism. After working with both I think the flux VAE has a slight edge in learning realistic textures but it's smaller than you might think, the Qwen VAE was overall very good in ablations and good at learning to produce a diverse set of styles.

[BoredPositron]

[flagged]

[dang]

> You can't be serious.

Please edit out swipes from your HN comments, as the guidelines request: https://news.ycombinator.com/newsguidelines.html.

Edit: your account has unfortunately been breaking the site guidelines like this in other places as well (e.g. https://news.ycombinator.com/item?id=48567675). Can you please fix this? I don't want to ban you, but we've already had to ask you this before.

[BoredPositron]

Kill the account dang.

[mattnewton]

Definitely encourage you to test the models. We tried to optimize for realistic focus and not over-sharpening, which leads to a "hyper" AI-look. It's hard to benchmark because people generally prefer sharp, saturated orangish pictures all else equal, but I believe these are bad shortcuts for the model to learn realism.

[BoredPositron]

Is my taste the problem, or am I simply holding it wrong? The qwen VAE's shortcomings are well documented, and Krea 2 produces the same blurry, airbrushed output as qwen image. Between the chaotic release and every interaction I've had with your team, I've grown to genuinely dislike your platform/company. Good luck.