Type | |
Stats | 1,031 |
Reviews | (222) |
Published | Jul 4, 2023 |
Base Model | |
Trigger Words | a 2koma of a girl sitting confused at a computer |
Hash | AutoV2 9ED6B8A06C |
Have you ever thought "Man I really need a LoRA to make confused girls sitting at computers"? No? Just me then?
I made this as an experiment to see if it was possible to make a good concept LoRA on just one image and the answer is "no". So I retrained with a second image made by the LoRA itself and the answer became - eh, kinda, fun to play with for a bit so I figured I'd share it.
The art style is very strict but it seems to work OK with character LoRAs. Slightly more reliable using DPM++ 2M Karras, at least on AOM2. Recommended LoRA strength 0.8 to 1.0. Note that the dataset had square images so using other aspect ratios might break things or add more panes.
If anyone wants to use this LoRA to make more images to train on for a better version feel free, no rights reserved whatsoever.
Original image created by the artist naavs.