This LoRA should perform well for any background graphics, with the optional help of an AMAZING pixel art WebUI add-on - <a target="_blank" rel="ugc" href="https://github.com/AUTOMATIC1111/stable-diffusion-webui-pixelization">pixelization</a>.If the pixelization shows ringing artifacts, please increase the pixel size, since this issue stems from the size of the image (the bigger image is and the smaller pixel size is, the more 'echoing' you could see with the pixelization attempt).That's how the add-on presents itself:<img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/e28d612b-575c-436f-874b-89e7273e7946/width=525/e28d612b-575c-436f-874b-89e7273e7946.jpeg" />The LoRAs were trained on my <a target="_blank" rel="ugc" href="https://civarchive.com/models/1494964?modelVersionId=1691229">ILL v0.1 "CLIP from Illustrious 2.0" infusion attempt</a> checkpoint as a base - they should perform well on most ILL, NoobAI or ever VPred variants. I didn't use any "people-y" LucasArts art for the training.v2This version was trained on the same set, as v1 and v0, but with improved (I hope) tagging, and also with regularisation images, tailored for 'style' models.I've used first two "space-y" images for the subsequent epochs of the LoRA for comparison purposes. :-)

Lucasarts Style

Actually second attempt to train it, because tagging was totally weird at first try. :)

Files with SHA256: 724af24f1d48f3b219c6becf0e205bbf81d59a2c9da3dc576119bb131507b224