FreeCustom is a novel tuning-free method for generating customized images with multi-concept composition. The method uses only one image per concept as input and introduces a new multi-reference self-attention (MRSA) mechanism and a weighted mask strategy to enable the generated image to access and focus more on the reference concepts. The MRSA mechanism leverages the context interaction of the input concept to capture the global context of the input images, allowing the method to rapidly generate high-fidelity images that align precisely with the text and maintain consistency with reference concepts without the need for training. The method is compared with existing approaches in terms of single-concept customization and multi-concept composition, and it is shown to outperform or perform on par with other training-based methods in terms of multi-concept composition and single-concept customization, but is simpler. The method is also shown to be robust and effective in generating high-quality images across diverse concepts. The method is applicable to various diffusion-based models and can be easily applied to other models. The method is evaluated using various metrics, including image similarity, image-text alignment, and image quality, and it is shown to achieve comparable results to other methods in terms of single-concept customization and exhibits noteworthy advantages when combining multiple concepts. The method is also shown to be effective in preserving the identity of the given concepts and generating images that align closely with the intended concepts. The method is also shown to be efficient in terms of time efficiency, as it does not require additional training. The method is evaluated through user studies and is shown to be superior in terms of image quality, identity fidelity, and text alignment. The method is also shown to be effective in applications such as appearance transfer and combining with existing methods. The method is supported by extensive experiments and is shown to be effective in various scenarios. The method is also shown to be flexible and robust, as it can directly work on different base models. The method is proposed to address the challenges of multi-concept customized composition and is shown to enable flexible combinations of different objects from various categories. The method is supported by a variety of references and is shown to be effective in various scenarios.FreeCustom is a novel tuning-free method for generating customized images with multi-concept composition. The method uses only one image per concept as input and introduces a new multi-reference self-attention (MRSA) mechanism and a weighted mask strategy to enable the generated image to access and focus more on the reference concepts. The MRSA mechanism leverages the context interaction of the input concept to capture the global context of the input images, allowing the method to rapidly generate high-fidelity images that align precisely with the text and maintain consistency with reference concepts without the need for training. The method is compared with existing approaches in terms of single-concept customization and multi-concept composition, and it is shown to outperform or perform on par with other training-based methods in terms of multi-concept composition and single-concept customization, but is simpler. The method is also shown to be robust and effective in generating high-quality images across diverse concepts. The method is applicable to various diffusion-based models and can be easily applied to other models. The method is evaluated using various metrics, including image similarity, image-text alignment, and image quality, and it is shown to achieve comparable results to other methods in terms of single-concept customization and exhibits noteworthy advantages when combining multiple concepts. The method is also shown to be effective in preserving the identity of the given concepts and generating images that align closely with the intended concepts. The method is also shown to be efficient in terms of time efficiency, as it does not require additional training. The method is evaluated through user studies and is shown to be superior in terms of image quality, identity fidelity, and text alignment. The method is also shown to be effective in applications such as appearance transfer and combining with existing methods. The method is supported by extensive experiments and is shown to be effective in various scenarios. The method is also shown to be flexible and robust, as it can directly work on different base models. The method is proposed to address the challenges of multi-concept customized composition and is shown to enable flexible combinations of different objects from various categories. The method is supported by a variety of references and is shown to be effective in various scenarios.