In the realm of artificial intelligence, data serves as a pivotal resource. Among these, real-world hyperspectral images (HSIs), bearing a broad spectrum of characteristics and diversity, are particularly valuable. However, the acquisition of HSIs can often be both financially burdensome and time-intensive. This results in a significant scarcity of HSI data, thereby hampering the progress and potential of deep learning applications that rely on HSIs. Current solutions aimed at remedying this shortage are largely inadequate, failing in generating a sufficient volume of diverse and reliable synthetic HSIs. In response to this, our study formulate a novel, generalized paradigm for HSI synthesis that initiates with unmixing, and follows by multi-source data fusion. Our approach pioneers the synthesis of hyperspectral data in the abundance domain, rather than the original, high-dimensional HSI space. We also incorporate multi-source data to augment the diversity of spatial distribution that the model can perceive and employ the unmixing concept to ensure the reliability of the spectral profiles of the synthesized HSIs. As a result, our approach is capable of generating a vast quantity of HSIs that cover a wide range of categories and scenes, closely mirroring realistic data. Our proposed generalized paradigm is instrumental in enhancing the reliability of the generative model's output and its applicability to real-world scenarios.