>>8945897
It's easier to talk about specific checkpoints and metadata rather than "the porn image trained models". I would download it and test if my suggestion works before putting it out here.

For a general approach you can look into Vectorscope or Diffusion-CG (different tools, same effect), these let you add a color or luminosity bias to your gens. Similar to post-production in photoshop, but since the model can already "see" this during generation it can adjust the scene to match. Like by removing light sources in a dark scene for example, or adding a sunset glow if you bias it towards orange, etc.

Test pic is genned on Lustify, same prompt and seed except different Vectorscope settings.