InstructIR is an image restoration model that uses human-written instructions to guide the restoration process. The model can handle various types and levels of image degradation, such as noise, blur, haze, and low-light conditions, and generalizes well in real-world scenarios. The model is trained using a large dataset of degraded and clean images, along with text prompts generated by GPT-4. The text prompts are encoded into embeddings using a sentence encoder, and the model is trained to understand and apply these instructions to restore images. The model uses a task routing mechanism to handle multiple restoration tasks simultaneously, allowing it to process different types of degradation efficiently. InstructIR achieves state-of-the-art results on several image restoration tasks, including denoising, deblurring, dehazing, and low-light enhancement. The model is efficient and can be trained on standard GPUs, making it suitable for real-world applications. The model's performance is evaluated on various benchmark datasets, and it outperforms previous methods in terms of image restoration quality and efficiency. InstructIR demonstrates the effectiveness of using text-based instructions to guide image restoration, providing a novel approach for text-guided image restoration and enhancement.InstructIR is an image restoration model that uses human-written instructions to guide the restoration process. The model can handle various types and levels of image degradation, such as noise, blur, haze, and low-light conditions, and generalizes well in real-world scenarios. The model is trained using a large dataset of degraded and clean images, along with text prompts generated by GPT-4. The text prompts are encoded into embeddings using a sentence encoder, and the model is trained to understand and apply these instructions to restore images. The model uses a task routing mechanism to handle multiple restoration tasks simultaneously, allowing it to process different types of degradation efficiently. InstructIR achieves state-of-the-art results on several image restoration tasks, including denoising, deblurring, dehazing, and low-light enhancement. The model is efficient and can be trained on standard GPUs, making it suitable for real-world applications. The model's performance is evaluated on various benchmark datasets, and it outperforms previous methods in terms of image restoration quality and efficiency. InstructIR demonstrates the effectiveness of using text-based instructions to guide image restoration, providing a novel approach for text-guided image restoration and enhancement.