Type | |
Stats | 3,594 55,004 |
Reviews | (281) |
Published | May 29, 2024 |
Base Model | |
Training | Steps: 375 Epochs: 10 |
Usage Tips | Clip Skip: 1 Strength: 0.8 |
Trigger Words | dark night backlit red light neon lighting |
Hash | AutoV2 857E8D9667 |
I wanted to give LoRA training feature a go, and I felt like I was missing "LowRA" from 1.5 so I used some stock photos that I thought would give a similar effect. I wasn't picky with my selection, and basically just took the first things I saw. There's a bias for cities, women and asian lettering.
This could probably be accomplished without LoRA at all, but I know it works and I just wanted to try recapping some buzz so I could play around with the feature some more.
Example images generated using datassRev3 in comfyui and a1111
DPM++ 2M Karras
20 steps, cfg 6
adetailer (face fix)
Upscaler: 4x_NMKD-Siax_200k
CNet Canny was used 0.01% for base image/noise because I didn't bother changing workflow.
Workflow on ComfyUI is a schizophrenic mess, im noob