I played a little bit with the image generation, and I'm not sure if there was something wrong with the site or was it was some weird setting, but it was Stable Diffusion 1.5 tier generation. Not that good :\
That’s not Janus Pro, its Janus Flow, an earlier DeepSeek model in the same development series, released last November.
The full model is 7 billion parameters, so I'm sure that's impacting the quality but as also pointed out in a different thread, apparently the maximum resolution is limited to 384x384.
You are using the old 1.3B model no wonder..
Yeah I tried it too on HuggingFace and was not impressed.