Alibaba has launched a new AI portrait image processing framework, UniPortrait, which can realize personalized processing of single-character and multi-character images, and supports style reference and free text description to generate images. UniPortrait uses a two-stage training scheme to maintain a high degree of lifelike facial features while supporting a wide range of facial editing functions, significantly improving the quality and efficiency of image generation. Its efficient customization capabilities and compatibility with existing generation control tools provide users with an unprecedented image personalization customization experience. The emergence of UniPortrait will undoubtedly push portrait image processing technology to new heights and bring more possibilities to various application scenarios in the future.
This framework can not only handle images of a single identity, but also enable high-quality personalization in multi-role scenarios. UniPortrait features the ability to maintain a high degree of realism in facial features and supports a wide range of facial editing capabilities, allowing users to even use free-form text descriptions to generate the images they want without the need for a fixed layout.
Multi-role consistency
During the training process, UniPortrait uses a carefully designed two-stage training program. The first phase focuses on the training of a single identity, while the second phase is fine-tuning multiple identities. Through this training method, UniPortrait outperforms existing methods in both single and multiple identity customization performance. Experimental results also show that it has good scalability and is generally compatible with existing generation control tools.
The launch of UniPortrait brings new possibilities for personalization of portrait images, especially in terms of free-form prompts and diverse layout generation. The research team demonstrated multiple examples of single and multiple identity personalization, showing the great potential of this framework for practical applications. In short, UniPortrait not only improves the quality of image generation, but also paves the way for various future application scenarios.
Product project entrance: https://top.aibase.com/tool/uniportrait
Trial entrance: https://huggingface.co/spaces/Junjie96/UniPortrait
Highlight:
Portrait is a new framework focused on single and multiple identity image personalization with high-quality facial feature preservation.
✍️ The framework consists of an ID embedding module and an ID routing module, achieving efficient customization through a two-stage training scheme.
UniPortrait provides rich possibilities for portrait personalization, supporting free text description and diverse layout generation.
All in all, UniPortrait provides users with a new image personalized customization experience with its powerful functions and convenient operations. It has great potential for future development and is worth looking forward to its application in more fields.