This model was trained with base FLUX DEV it can be create coherent images with prompt engineering and trying different weight ratios, or in order to create the images you want you need to put them in Comfy as reference images and let this Lora enhance your image, so try different denoise rates or Lora weight ratios and enjoy.