Updated: Mar 24, 2026
character
Wanted to test a realistic character with this model type, I must say, I love the result!
Her gloves and sleeves can and will be fucked up, 50%-50%.
Trained on IL XL 1.1
Read the rules!!! It's down in the suggesteds.
Low-Rank Hadamard Product (LoHa) is a Parameter-Efficient Fine-Tuning (PEFT) method designed to improve upon Low-Rank Adaptation (LoRA) by increasing the expressivity of model updates without requiring additional trainable parameters.
LoHa achieves this by replacing the standard matrix product used in LoRA with the Hadamard product (element-wise multiplication) to compute weight changes. (I don't know what this means, maybe you do.)
File size caught me off-guard, a LoRa with the same weights would be 106 mb, no matter, result is superb. It is double the layers if I understand correctly. Works the same way as a regular LoRa. A1111, Forge and Comfy has support since 2023.
Outfit: asymmetrical clothes, capelet, black jacket, purple shirt, single fingerless glove, sleeve gloves, arm strap, pants, multiple thigh straps, knee boots
Generated with DPM++ 2M SDE - Beta, 25 steps, CFG 4.
I use No embeddings, NO negatives and NO quality tags at all! Only the style and medium tags.
Recommended Resolutions: (SDXL standards) - Pony/IL too
640 x 1536 / 1536 x 640
768 x 1344 / 1344 x 768
832 x 1216 / 1216 x 832
896 x 1152 / 1152 x 896
1024 x 1024