Palmprint has recently received increasing attention for biometric recognition due to its high convenience and low invasiveness, while the scarcity of large-scale real public palmprint databases significantly limits the development of palmprint recognition. Synthetic image generation provides a new way for establishing a large-scale palmprint image database. However, most existing methods simulate geometric curves to generate completely virtual palmprint images, which have a large gap from the real samples. In this paper, we propose an identity-traceable stylized palmprint image generation method (IdTrPalm) based on a conditional diffusion model by transferring the style factors from a real palmprint image into another real one. Moreover, we collect massive “one-palm one-capture” real-world palmprint images captured under diverse imaging conditions from the Internet. With the diverse real palmprint samples and the proposed generation method, we generate a large number of realistic palmprint images with traceable palm identities, establishing a large-scale identity-traceable palmprint image database. Extensive experimental results demonstrate that our method generates high-quality palmprint images with superior visual realism and significant performance promotion in palmprint recognition.