OneRestore is a universal restoration framework designed to address complex, composite degradations in images, such as low light, haze, rain, and snow. The framework introduces a transformer-based model that integrates scene descriptors, derived from either manual text embeddings or automatic visual attribute extractions, to enable adaptive and controllable image restoration. A key innovation is the Scene Descriptor-guided Transformer Block (SDTB), which enhances the model's ability to distinguish between different degradation types. The framework also incorporates a composite degradation restoration loss, using additional degraded images as negative samples to strengthen model constraints. Experimental results on synthetic and real-world datasets demonstrate that OneRestore outperforms existing methods in handling complex degradations, achieving state-of-the-art performance. The model's versatility allows it to handle multiple degradation types simultaneously, making it a robust solution for real-world image restoration tasks. The framework is evaluated against various state-of-the-art methods, showing its effectiveness in restoring images under diverse degradation conditions. The study highlights the importance of integrating scene descriptors and advanced loss functions to improve the accuracy and controllability of image restoration.OneRestore is a universal restoration framework designed to address complex, composite degradations in images, such as low light, haze, rain, and snow. The framework introduces a transformer-based model that integrates scene descriptors, derived from either manual text embeddings or automatic visual attribute extractions, to enable adaptive and controllable image restoration. A key innovation is the Scene Descriptor-guided Transformer Block (SDTB), which enhances the model's ability to distinguish between different degradation types. The framework also incorporates a composite degradation restoration loss, using additional degraded images as negative samples to strengthen model constraints. Experimental results on synthetic and real-world datasets demonstrate that OneRestore outperforms existing methods in handling complex degradations, achieving state-of-the-art performance. The model's versatility allows it to handle multiple degradation types simultaneously, making it a robust solution for real-world image restoration tasks. The framework is evaluated against various state-of-the-art methods, showing its effectiveness in restoring images under diverse degradation conditions. The study highlights the importance of integrating scene descriptors and advanced loss functions to improve the accuracy and controllability of image restoration.