Z-Image already has a fantastic digital painting style, natively. In fact, I thought I wouldn't be able to achieve this LoRA training at all - the innate Z-Image aesthetic (so close already to this style) consistently overwhelmed what I was aiming for after a handful of training runs. But with some lucky merges between different sets, I got something decent and consistent here. A bit glossier, with more glow. More ambiance and less hyper-detailing and near-photographic looks.

I call it 'Digital Painting', and it is suitable as an all-around illustration model, but it's truly a successor to my Flux LoRA Eldritch Fantasy. Trained on the same dataset and fantasy subjects are a primary interest for me, which you can see in my sample output. Sadly, Z-Image doesn't know quite as many fantasy characters as Flux, making it a bit less suitable as a fantasy image generator (and I'm not training thousands of images to teach new concepts, just dozens to teach style) but it's still pretty darn good.

Prompt for "Digital Painting ..." and you should be good to go.

Downsides: not enough variety of faces, without careful prompting. Sometimes still just veers into photorealism or Z-images default aesthetic.