I've been trying to train a lora to achieve better dark images for quite a while, but was never satisfied with the results. The way SD works isn't ideal to be able to achieve this. This version, however, does produce good results. It works by combining some of the tags below and the weight of the lora. I recommend combining multiple tags from the list below. Note that the end result ultimately depends on the base model used.

dark theme: general darker image

dim lighting: darker image with dim light source

black theme: darken image based on making things towards the black color, but it can lead to black and white results if there is no color tags in the prompt

silhouette: darkens the character and tens to generate a rim light. I recommend using with the tag "solo" as it can generate silhouettes of others in the background

dark background: darkens the background. It can generate a black background if none is described in the prompt

dark room: similar to "dark background"

black background: having a black background darkens the overall image

backlighting: generate the effect of light coming from behind the character, and it darken the character as well. I also tried to train with other tags, which you can try, but don't seem very effective: underlighting, sidelighting, uplighting, overlighting. Maybe for a future version I'll revisit this

Other tags you can play with: shadow, shade, red light, blue light, red theme, blue theme.

Recommended quality tags: score_9, score_8_up, score_7_up, score_6_up. Optional: source_anime

Recommended negative tags: score_4, score_5. Optional: 3d, lips, realistic