Converts an image of a person, animal, fantasy creature, etc. into a retro character sprite.

The keyword to activate the model is M4k3spr1t3, which you can use as a verb before a subject in the input image; for example, if there is a photo of a man and a woman standing side-by-side, then you would use "M4k3spr1t3 the man" to make a sprite of just him but would use "M4k3spr1t3 the people" or "M4k3spr1t3 the man and woman" to include both of them.

M4k3spr1t3 <character>
M4k3spr1t3 the man
M4k3spr1t3 the woman
M4k3spr1t3 the animal
M4k3spr1t3 the creature

It is not always required, but I also recommend including "white background" and "full body" as part of the prompt, which will help prevent the subject from being out of frame or creating a pixelized background instead of removing it.

M4k3spr1t3 the person, full body, white background

The model will attempt to preserve their pose and orientation as it exists in the original image unless directed otherwise; it is capable of modifying subjects in the process if you want to include elements like "facing the viewer," "looking over shoulder," etc.

I have attempted to make the output a uniform style, however I've found that higher resolutions tend to result in the pixel style effect being diminished. If you generate at a resolution of 256 or 512, you are almost guaranteed to generate the style depicted in the sample images; if you generate at a greater resolution like 1024, then there is the possibility that it produces a higher resolution pixel art, reminiscent of 32-bit pixel art.

You can enforce the resolution by using the "Resize and Pad" node on the input image and disabling the upscale step in the default Klein workflow's subgraph (the first node, which upscales the image to one megapixel).

California AB 2013 Training Data Disclosure

• This LoRA was fine-tuned using visual data consisting of still images that were publicly available online. The training data includes copyrighted material owned by third parties. No training data was licensed or purchased. This LoRA is provided for non-commercial use only under the terms of its distribution.

• The dataset consists of 155 image pairs (310 images total). Each pair was created by taking a sprite consistent with the desired style and locating an image of the same subject in a different style. Data was collected and generated between approximately 2024 and 2026.

• Image data was processed through standard resizing, cropping, normalization, and labeling steps. Synthetic images and video were not included as part of the training dataset.

• This model is intended for non-commercial, experimental, and educational use. Generated outputs may reflect copyrighted visual styles or themes associated with the underlying training data. Users are responsible for ensuring compliance with applicable copyright law, other intellectual property laws, and all other applicable laws.