If this is actually it, it's nowhere near DALL-E quality, and even further from Imagen.

Even taking into account the limited resolution, this is more like SD1.

Quality notwithstanding, I don't think it even competes against the SOTA for understanding either - I ran some comparisons against Flux and was unimpressed.

https://imgur.com/a/nn9c0hB

Notably, the benchmarks don't compare to existing models on image quality, which is hard to measure outside of the Image Arena-style tests.

The release is more about the multimodal captioning which is an objective improvement. I'm not a fan of the submission title.

no someone forked an older janus model. pro is not released yet

no someone forked an older janus model. pro is not released yet