This paper introduces “StainNorm,” a novel solution for addressing the challenge of stain colour normalization in histopathology images. Stain colour inconsistency presents a significant hurdle to accurate analysis in digital pathology. To tackle this issue, I present an innovative CycleGAN architecture with a 1 × 1 kernel, enabling efficient yet robust stain colour normalization. StainNorm is rigorously evaluated across diverse datasets, including Stardist’s training data (TNBC-TMA and MoNuSeg 2018), an in-house whole-slide image (WSI) dataset, and pathologist-curated TNBC images for validation. Custom models are trained for each use case. A key breakthrough in StainNorm is the adoption of a streamlined training process, leveraging multiphase training with an identity transform on the source set and gradually incorporating colour jitter using a task-based curriculum learning strategy. This innovation reduces training time by 40 times, allowing training on systems with limited resources. StainNorm’s efficacy is demonstrated through comprehensive comparisons against established techniques, outperforming state-of-the-art deep learning methods in normalizing images to the target distribution, in structural similarity index measure (SSIM), Jensen-Shannon (JS) distance and computational complexity. StainNorm excels in normalizing background areas which are often poorly addressed by StainNet, the previous state of the art (SOTA). While deep learning methods often demand substantial resources, StainNorm’s efficiency bridges this gap by minimising training time, making it practical for integration into histopathology laboratories.