Scholarlink

<p>We introduce <i>Gorgeous</i>, a diffusion-based generative method that redefines digital makeup application by enabling the generation of creative makeup designs through image prompts. Unlike conventional makeup transfer techniques that primarily replicate existing styles, <i>Gorgeous</i> is the first to empower users to incorporate narrative elements into makeup ideation via image-based prompts. This approach allows for the creation of makeup concepts that visually convey user expressions, offering innovative and imaginative makeup ideas for real-world application. To achieve this, <i>Gorgeous</i> establishes a foundational learning framework, ensuring that the model first comprehends “what makeup is” before integrating narrative elements. A pseudo-pairing strategy, leveraging a face parsing and content-style disentangling network, effectively addresses challenges associated with unpaired data, enabling the model to be trained on bare-face images. Users can input reference images that represent their conceptual ideas (e.g., fire), from which <i>Gorgeous</i> extracts contextual embeddings to guide a novel makeup inpainting algorithm. This process facilitates the generation of creative, narrative-driven makeup designs tailored to specific facial regions. Comprehensive experiments validate the effectiveness of <i>Gorgeous</i>, demonstrating its potential to establish a new paradigm in digital makeup artistry and application. Code released at: <a href="https://github.com/JiaWeiSii/gorgeous">https://github.com/JiaWeiSii/gorgeous</a>.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Gorgeous: Creating narrative-driven makeup ideas via image prompts

  • 期刊论文

  • 发表日期：

    2025年5月5日

查看全文
  • Jia Wei Sii,
  • Chee Seng Chan

摘要

We introduce Gorgeous, a diffusion-based generative method that redefines digital makeup application by enabling the generation of creative makeup designs through image prompts. Unlike conventional makeup transfer techniques that primarily replicate existing styles, Gorgeous is the first to empower users to incorporate narrative elements into makeup ideation via image-based prompts. This approach allows for the creation of makeup concepts that visually convey user expressions, offering innovative and imaginative makeup ideas for real-world application. To achieve this, Gorgeous establishes a foundational learning framework, ensuring that the model first comprehends “what makeup is” before integrating narrative elements. A pseudo-pairing strategy, leveraging a face parsing and content-style disentangling network, effectively addresses challenges associated with unpaired data, enabling the model to be trained on bare-face images. Users can input reference images that represent their conceptual ideas (e.g., fire), from which Gorgeous extracts contextual embeddings to guide a novel makeup inpainting algorithm. This process facilitates the generation of creative, narrative-driven makeup designs tailored to specific facial regions. Comprehensive experiments validate the effectiveness of Gorgeous, demonstrating its potential to establish a new paradigm in digital makeup artistry and application. Code released at: https://github.com/JiaWeiSii/gorgeous.