One of my preferred test prompts when training LoRAs is "colorful cityscape at night." It usually generates some of the most uniform results at early epochs and its generations are usually among the last to be affected stylistically at later epochs. So when I'm trying to create a style LoRA without a trigger word, it's a good measure of the training having a general effect.

Also, I like to save interesting LoRA training epoch sample images for possible future inspiration, use, or posting. Both of these facts can be seen in my work here on Civitai.

So I've amassed a small collection of "colorful cityscape at night" images in different styles. For this model, I've trained the tags "colorful," "cityscape," "night," and some related concepts using some of those images to bring some variety to art generated using those and related prompts.