IMPORTANT:
You must use the term xyzvoidglitch to trigger the LoRA, here's an example:
xyzvoidglitch, <lora:xyzVoidGlitch:1> solo man sitting on a livingroom chair, striped purple and red shirt, holding, nasa backrooms, indoors, cinematic, dynamic, night
Overview:
Trained on 30 manually glitched images, this LoRA produces imagery that takes the silhouette of figures and glitches them to hell in a variety of ways.
Recommended Settings:
LoRA Weight: Full strength of 1 works great.
Models: v1-5-pruned-emaonly (trained on), ProtoGen_X3.4-pruned-fp16, realisticVisionV20_20, f222
Clip skip: 1 or 2
Resolution: Training data was 768x768 and all standard resolutions up to 512 x 1024 seem fine, 1024 x 1024 is where quality seems to fall off.
Sampler: DPM++ 2M Karras
Steps: 25-50
CFG: 6-10
Example Prompt:
Model: v1-5-pruned-emaonly
Steps: 30, Sampler: DPM++ 2S a Karras, CFG scale: 9, Seed: 3478988281, Size: 512x896
Posetive: xyzvoidglitch, <lora:xyzVoidGlitch:1>, black and white street photography, (maelstrom:1.2), conflict, body, skin, footage, indoors, ilford hp5, cinematic lighting, dynamic composition, night, contrast, cinestill 800t, cinematic
Negative: n/a
A note on image quality:
The original images were old 90's film and point and shoot, many with flash and a fair bit of noise, this was intentional on my part, just be aware that getting something super clean and noise free may be a bit more of a chore than usual. The point is to be grungy and nostalgic, that said I think the quality is still pretty great.