
At the risk of cannibalizing my downloads: if file size is a concern for you, I highly recommend trying out my LoKr with the same training data set, since it is tiny (~7MB): https://civitai.com/models/129177/desi-coffee-sdxl-lokr-tiny-sized-lora-for-indian-faces-skin-tones-experimental
This is an SDXL version of the Desi Coffee LoRA to generate Indian looking faces and skintone.
The same trigger words work as before:
-
desi (general South Asian tone)
-
desilatte (light tone)
-
desimocha (medium tones)
-
desiespresso (dark tones)
V2 produces much better outputs than V1, though it was trained on the same base data. Use the 2.0 version unless there's a reason for using 1.0.
Other settings:
-
CFG: 6-8
-
Steps: 20-30
-
Strength: 0.6-1
-
Sampler: euler a
Excuse the large size, need to still figure out what settings work best. while making SDXL LoRAs
描述:
Updated version of the LoRA that is smaller, but produces much better images. The prompts remain the same: desilatte, desimocha, and desiespresso
训练词语: desilatte,desimocha,desiespresso
名称: DesiCoffee_SDXL_20_LoRA.safetensors
大小 (KB): 223113
类型: Model
Pickle 扫描结果: Success
Pickle 扫描信息: No Pickle imports
病毒扫描结果: Success