**StyleShot: A Snapshot on Any Style** **Authors:** Junyao Gao, Yanchen Liu, Yanan Sun, Yinhao Tang, Yanhong Zeng, Kai Chen, Cairong Zhao **Institution:** Tongji University, Shanghai AI Laboratory **Project Page:** https://styleshot.github.io/ **Abstract:** This paper introduces StyleShot, a method for generalized style transfer without test-time tuning. StyleShot achieves this through a style-aware encoder and a well-organized dataset called StyleGallery. The style-aware encoder is designed to extract expressive style representations, while StyleGallery ensures generalization. A content-fusion encoder enhances image-driven style transfer. StyleShot can mimic various styles, including 3D, flat, abstract, and fine-grained styles, without test-time tuning. Extensive experiments show that StyleShot outperforms existing methods in a wide range of styles. **Contributions:** - Proposes StyleShot for generalized style transfer. - Introduces a style-aware encoder and a content-fusion encoder. - Develops a comprehensive style benchmark, StyleBench, covering 73 distinct styles. - Conducts extensive evaluations to validate StyleShot's effectiveness and superiority. **Method:** - **Style-Aware Encoder:** Extracts rich and expressive style embeddings from reference images using multi-scale patch partitioning and Transformer Blocks. - **Content-Fusion Encoder:** Integrates content and style information to enhance style transfer performance. - **StyleGallery:** A balanced dataset with diverse styles to improve model generalization. - **De-stylization:** Removes style-related descriptions from text prompts to improve style extraction. **Experiments:** - **Qualitative Results:** StyleShot effectively captures and transfers a broad spectrum of styles, including fine-grained details. - **Quantitative Results:** StyleShot achieves superior performance in text and image alignment compared to existing methods. - **Ablation Studies:** Demonstrates the effectiveness of each component in StyleShot. **Conclusion:** StyleShot is a powerful tool for style transfer, capable of accurately identifying and transferring styles from any reference image without test-time tuning. It excels in both text-driven and image-driven style transfer tasks, demonstrating superior performance and robustness across various styles.**StyleShot: A Snapshot on Any Style** **Authors:** Junyao Gao, Yanchen Liu, Yanan Sun, Yinhao Tang, Yanhong Zeng, Kai Chen, Cairong Zhao **Institution:** Tongji University, Shanghai AI Laboratory **Project Page:** https://styleshot.github.io/ **Abstract:** This paper introduces StyleShot, a method for generalized style transfer without test-time tuning. StyleShot achieves this through a style-aware encoder and a well-organized dataset called StyleGallery. The style-aware encoder is designed to extract expressive style representations, while StyleGallery ensures generalization. A content-fusion encoder enhances image-driven style transfer. StyleShot can mimic various styles, including 3D, flat, abstract, and fine-grained styles, without test-time tuning. Extensive experiments show that StyleShot outperforms existing methods in a wide range of styles. **Contributions:** - Proposes StyleShot for generalized style transfer. - Introduces a style-aware encoder and a content-fusion encoder. - Develops a comprehensive style benchmark, StyleBench, covering 73 distinct styles. - Conducts extensive evaluations to validate StyleShot's effectiveness and superiority. **Method:** - **Style-Aware Encoder:** Extracts rich and expressive style embeddings from reference images using multi-scale patch partitioning and Transformer Blocks. - **Content-Fusion Encoder:** Integrates content and style information to enhance style transfer performance. - **StyleGallery:** A balanced dataset with diverse styles to improve model generalization. - **De-stylization:** Removes style-related descriptions from text prompts to improve style extraction. **Experiments:** - **Qualitative Results:** StyleShot effectively captures and transfers a broad spectrum of styles, including fine-grained details. - **Quantitative Results:** StyleShot achieves superior performance in text and image alignment compared to existing methods. - **Ablation Studies:** Demonstrates the effectiveness of each component in StyleShot. **Conclusion:** StyleShot is a powerful tool for style transfer, capable of accurately identifying and transferring styles from any reference image without test-time tuning. It excels in both text-driven and image-driven style transfer tasks, demonstrating superior performance and robustness across various styles.