BeautyGlow is a framework that uses a reversible generative network called Glow to transfer makeup from a reference image to a target image. It proposes decomposing the latent vectors derived from Glow into makeup and non-makeup latent vectors using a new transformation matrix and loss function. This allows for on-demand makeup adjustment by interpolating the latent space. Experimental results show BeautyGlow is comparable to state-of-the-art methods while also enabling control over makeup extent from light to heavy using the latent vectors.