LyliaEngine/UUDC

Urban_Liminal_Underground_Construction_Style_Background

Prompt

score_9, score_8_up, score_7_up, score_6_up, BREAK UUDC, indoors, horror \(theme\), dark, scenery, abandoned, underground, room, bathroom, concrete walls, tile walls, tile floor, fluorescent light, dim lighting, decayed, damaged walls, neglected, no humans, shadows, architecture, urban decay, bathroom stalls, broken tiles, rubble, drain cover, floor drain, shelf <lora:[Urban Liminal] Underground Construction Style Background_epoch_10:0.6>

Negative Prompt

pony, gaping, censored, furry, 3d, aidxlv05_neg, realism, realistic, deformed hands, multiple fingers, (scenery:1.2), (no humans:1.2), architecture, skirt

Prompt

score_9, score_8_up, score_7_up, score_6_up, source anime BREAK 1girl, blonde hair, ponytail, white shirt, small breast, denim jacket, blue skirt, leather boots, belt, walking, scared, looking to the side, worried, (cowboy shot:1.4), (dutch angle:1.8), BREAK UUDC, indoors, horror \(theme\), dark, abandoned, underground, corridor, hallway, tunnel, concrete walls, peeling paint, dirty floor, cracked floor, dim lighting, decayed, damaged walls, neglected, concrete floor, shadows, urban decay<lora:[Urban Liminal] Underground Construction Style Background_epoch_10:0.6>

Negative Prompt

pony, gaping, censored, furry, 3d, aidxlv05_neg, realism, realistic, deformed hands, multiple fingers, (scenery:1.2), (no humans:1.2), architecture, (wide shot:1.2)

Model description

Please help me do more LORA's by tipping Buzz on the model or by posting images of it I'd greatly appreciate it c:

Trained from screenshots I took in Gmod from the map ttt_slender_v3_fix https://steamcommunity.com/sharedfiles/filedetails/?id=143546559

This is by far the messiest model I've made for backgrounds. It works, but some trial and error is necessary. Here are some things I've noticed before releasing it... It doesn't have a specific trigger, but UUDC appears at the beginning of all the images in the dataset.

Weight: It works well between 0.6 and 0.8. It also works at 1.0, but keep in mind that the higher the weight, the less versatile the model will be. Generally, for using characters in the background, 0.6 to 0.8 works well, and for just the background, between 0.8 and 1.0.

Hi-Res Fix: When there is a character present, it tends to be more stable. However, when it's just the background, the model is highly unstable with Hi-Res Fix, even with low levels of denoising. It is recommended to do the Upscale in Img2Img using the DPM++ 2M Karras sampler with a not too high denoising between 0.35-0.5; even 0.6 can work well... instead of using Hi-Res Fix. Another thing is that when using Img2Img, be mindful of how much weight is given to the LoRA; for cleaner results, 0.6 is fine. The higher you go, the more "realism" it adds to the image, which also contributes to a creepier aesthetic.

Sometimes it’s difficult to get a character to appear in the scene, so it’s recommended to put the following in the negative tags:

scenery, no humans, architecture

Important tags for the model:

indoors, horror \(theme\), dark, abandoned, underground, tunnel, concrete walls, peeling paint, dirty floor, cracked floor, fluorescent light, dim lighting, decayed, damaged walls, neglected, concrete floor, shadows, urban decay

General tags to use with the model:

pillar, black stain on wall, industrial pipe, bathroom, bathroom stalls, dirty walls, peeling paint, dirty floor, graffiti, broken tiles, garbage, ceiling light, toilet, drain cover, floor drain, boarded door, rubble, broken door, pile of debris

Other LORAS used on the preview pics:

Albinauric Monster: https://civitai.com/models/395351?modelVersionId=440953 Strange World: https://civitai.com/models/499581?modelVersionId=555311

Source

https://civitai.com/models/533035?modelVersionId=592445

Credit

https://civitai.com/user/Joseidazu

Trigger words

You should use UUDC to trigger the image generation.

Download model

Weights for this model are available in Safetensors format.

Download them in the Files & versions tab.