Be sure to remove the 'easy negative' from prompt settings. It works against the fact that these are supposed to be noisy images.

The effect is sometimes subtle, sometimes stark. Trained on a bunch of high-grain photos with a film-style reduced dynamic range (particularly with crushed blacks/shadows) this model brings an aesthetic that may feel pointless to some, and quite charming or romantic to others. Consider myself in the second camp (why I took the time to train it) but no hard feelings if it's not your thing. There's plenty of other style on this site for you.

In my local setup, I run with Z-Image Turbo and just keep the LoRA strength at 1.0 and I find no special prompt needed. However, on TensorArt, I find I prefer to reinforce the style with "grainy analog photo" and maybe bump the LoRA strength up to 1.1 or 1.2.