Abstract
Most existing illumination-editing approaches fail to simultaneously providecustomized control of light effects and preserve content integrity. This makesthem less effective for practical lighting stylization requirements, especiallyin the challenging task of transferring complex light effects from a referenceimage to a user-specified target image. To address this problem, we proposeTransLight, a novel framework that enables high-fidelity and high-freedomtransfer of light effects. Extracting the light effect from the reference imageis the most critical and challenging step in our method. The difficulty lies inthe complex geometric structure features embedded in light effects that arehighly coupled with content in real-world scenarios. To achieve this, we firstpresent Generative Decoupling, where two fine-tuned diffusion models are usedto accurately separate image content and light effects, generating a newlycurated, million-scale dataset of image-content-light triplets. Then, we employIC-Light as the generative model and train our model with our triplets,injecting the reference lighting image as an additional conditioning signal.The resulting TransLight model enables customized and natural transfer ofdiverse light effects. Notably, by thoroughly disentangling light effects fromreference images, our generative decoupling strategy endows TransLight withhighly flexible illumination control. Experimental results establish TransLightas the first method to successfully transfer light effects across disparateimages, delivering more customized illumination control than existingtechniques and charting new directions for research in illuminationharmonization and editing.