Trained on 16 images from the new Midjourney Niji 5 model, using the standard low VRAM presets, 100 steps per image. Seems to give the image a little bit more pizazz and details, and seems to adhere to the prompt better, but thats just the opinion of an ape with a keyboard! It came out quite strong, so recommended to go with weightings between 0.2 and 0.6 to get decent results out of it. LoRA was trained with JAFA Mix v1 as the base, but sample images were made with Counterfeit and MeinaMix.

Files with SHA256: 733d61ea52bd3828934de9413b2dd844034ca1c93bf77907f8dd76ea46da706d