XUPLX (UglyPeopleLoRA) is intended to break up the monotony of ZImage Turbo output by forcing interesting compositions, character designs, camera angles, action and other attributes into the output image by using keywords as a prompt.

XUPLX was trained with "mood-tags" (see the prompts in the images). Within the workflow, the LoRA is activated by "moody" keywords; the mood image is generated within the first couple of steps with XUPLX, and then the rest of the generation is processed by a second Ksampler with a proper natural language prompt - without the LoRA.

XUPLX_UglyPeopleLoRA can be used on it's own (see "After and Before" images), or within the workflow.

MY LATEST WORKFLOW 23 DEC 2025 (Article coming soon)

https://civarchive.com/models/2245448/zimage-edit-upscale-merge-workflow

Use of the "XUPLX" trigger is not necessary - until it is. Z-Image-Turbo has weird behavior with certain hyphenated words (ie. "art-house" will stubbornly prioritize generating Kimonos over the generating your prompt).

NOTE: You can use "Black and white", "cigarette" or "smoking" in your prompt with XUPLX to see it in action (the LoRA reacts strongly).

DON'T BUY ME A COFFEE!!

Download ExportGPT - Media & Chat Export Toolkit from the Chrome Web Store. The chrome extension works with ChatGPT and Sora to export assets quickly. I update it often, it has 5 stars, and it's free forever!

Visit exportgpt.org to see it in action.