HAV0X1014/Chroma-KF-Bat-Eared-Fox-LoRA

Bat Eared Fox (Kemono Friends)

Prompt

bat_eared_fox_kemono_friends walking through a datacenter alone, she is looking around at the servers and small blinking lights in the server racks, holding a clipboard. she has a look of wonder as she curiously takes the sights of the dim servers around her.\n\nhigh quality, good detail, 1girl, full body, detailed background, good hands, masterpiece, anime style

Negative Prompt

low quality, bad anatomy, poor quality, bad image, low resolution, artefact, multiple tails, two tails

Prompt

a young woman sitting in a coffee shop, holding a coffee cup, sitting by the window during a heavy rainstorm. She is looking at the viewer, gently smiling\n\nhigh quality, good detail, 1girl, full body, detailed background, good hands, masterpiece, bat_eared_fox_kemono_friends

Negative Prompt

low quality, bad anatomy, poor quality, bad image, low resolution, artefact, multiple tails, two tails

Model description

The LoRAs in the "old version" folder are the old 2200 and 2000 step trained LoRAs at 1e-4 learning rate. The 4000 and 3500 step "battyV2" LoRAs are the newer, better trained ones trained at 2.5e-5 learning rate. I have had a better/more consistent experience using the new 4000 step LoRA. I also included the 3500 step checkpoint just in case it works better than the 4000 step one.

The rest of this readme is mostly the same.

A LoRA I trained on 64 images of Bat Eared Fox using ai-toolkit. I trained it to 2200 steps, however I think the 2000 step checkpoint is better/more consistent. I used too high of a learning rate (1e-4) on this one because I just wanted to see if training this would work at all.

I had to rent a 5090 off of vast.ai to train this because my stupid 7900 xtx won't train anything because bitsandbytes doesnt work right. It only cost me 75 cents out of the 5 dollars of credit I loaded in so I'm pretty happy with the cost this took.

trigger word is "bat_eared_fox_kemono_friends". Images were captioned by me by hand with a mix of danbooru tags and natural language.

Trained on Chroma v45 non detail calibrated, just the normal one.

Trigger words

You should use bat_eared_fox_kemono_friends to trigger the image generation.

Download model

Download them in the Files & versions tab.