A Z-Image Turbo LoRA trained on De-distilled model for reproducing Pup Play mask/hood. Even though dataset was composed of photos only, can be used for different styles, pretty versatile in that regard.

Don't forget to describe colors of mask and optionally material.

DON'T describe hairstyle since it messes the mask details.

One side effect - makes harnesses more realistic, since they were present in dataset.

Each image in showcase was generated with ComfyUI so it contains a workflow. For finer details to really shine - use higher base resolution (~1536) and second pass of KSampler with denoising 0.3 and small upscale.