Stable Diffusion Reference Only
stable-diffusion-reference-only is an image-to-image adaptation of Stable Diffusion designed for secondary painting tasks, including automatic coloring of line art, anime character remixing, and style transfer. Built on the reference-only mechanism, it allows users to upload a reference image to guide the color palette, style, and character features of a new generation based on a provided blueprint or line art. The software features an integrated Gradio web interface for easy local deployment or remote usage via Hugging Face Spaces and Google Colab. It includes specific utilities such as character segmentation to remove backgrounds, automatic line art extraction from color drawings, and color inversion to prepare blueprints. The model has been trained extensively on anime datasets at 512 resolution to produce high-quality results for fan art and concept design. Installation is supported via Python virtual environments like Conda, with compatibility for PyTorch and specific versions of the Diffusers library. T