Sigh. It's still a Pony. At this point, maybe restarting from scratch is a good idea. Try, for example, merging one of the latest BigLove checkpoints with ChlorideV1 using CosineA and all default settings. You'll be surprised at the result: BigLove "photo" series are broken without DMD2, but an injection of ChlorideV1 (use as Model B) with CosineA magically fixes it. With this method, you can merge things that look broken. With Cosine A, the first model (Model A) sets the structure (so poses and prompt adherence), while Model B is responsible for details.

The backgrounds, clothing and skin textures are all really nice, but the faces seem too cartoonish and not realistic enough to match everything else. Kinda feels/looks like a very early Illustrious semi-real model in that regard. Hands are absolutely outstanding imo, but anatomy seems a bit too exaggerated sometimes. Prompt adherence seems good and LoRA adherence seems fairly decent as well (not as good as test_V1 and Hands_Experiment v1.0, but better than Prompt4Pleasure v3.0).

Other than the faces, my biggest critique about NameGoHere_ v1.0 is the contrast seems really off, almost like it doesn't have a VAE baked-in (though I'm almost certain it does). The lighting, depth and reflectivity are all really nice, but the contrast just seems really low, like it's faded and washed-out (conversely, test_V1 might be just a bit too strong with its contrast, though not as extreme). If you're using an HDR monitor, it's very likely you wouldn't even notice this, but on a standard display, it's quite apparent for all results.

Tldr; Has unrealistic faces and very poor, washed-out contrast.