Digital Salon: An AI and Physics-Driven Tool for 3D Hair Grooming and Simulation

'2*, '3*, 1*†,
1, 1, 4,5, 6, 3, 7, 1, 2
1Adobe Research, 2Yale University, 3KAUST, 4Max Planck Institute for Intelligent Systems, 5ETH Zürich, 6Kiel University, 7Nanjing University of Science and Technology
* Equal contribution    † Corresponding author   ' Previous Adobe Research Intern

Abstract

Digital Salon is a revolutionary tool for 3D hair grooming and simulation, integrating AI and physics-based algorithms to simplify hair modeling, animation, and rendering. It enables users to generate detailed hairstyles through natural language descriptions. This tool combines text-based hair creation, interactive editing, and high-quality rendering into a unified pipeline. Digital Salon's innovative real-time simulation and versatile capabilities allow for the dynamic representation of various hair types and interactions with support for 10,000 to 80,000 strands, making advanced hair design accessible to users of all skill levels and dramatically streamlining the creative process in digital media.

Demos

All demos were run on either a laptop equipped with an NVIDIA GeForce RTX 4090 Laptop GPU or a desktop with an NVIDIA GeForce RTX 2080 Ti. Specifically, all computations, except for AI-image rendering, utilized 1 to 2 GB of GPU memory, depending on the size of the hair.

Text-guided Hair Generation

Our system facilitates 3D hair generation through text-based input. Upon entering a prompt, the system will provide three distinct hairstyles with variations. Users can select any of the generated options, and our system will automatically load the corresponding 3D data for further interaction.

"I want a Buzzed Cut with curly top."

"I changed my mind, I want voluminous, medium-length with waves. The curls are uniformly distributed and look dense and textured."

"Can you tie the curly hair to double ponytails?"

Interactive Hair Grooming

Once a hairstyle is either loaded or generated, users can further refine it through interactive grooming. Our system supports procedural strand growth, enabling the addition of interesting facial hair like beards. Additionally, it offers procedural strand cutting to remove unwanted parts. The entire process is interactive and simulatable, providing users with real-time feedback and a realistic experience.

Real-time Hair Simulation

Our system also supports real-time hair simulation with physically plausible dynamics. The driving force can be set by numeric values, by dragging the head mesh, or through text prompts.

"blow some wind?"
"create strong wind?"

AI-Rendered Images

Leveraging advanced image generation models, our system facilitates the creation of high-quality images, where the simply shaded head and hair serves as a structural control and some additional text prompts are introduced for supplementary elements not exist in the original rendering (e.g., cloth and background). The images generated by our system maintain a close alignment with the original hairstyle and are rendered in a photorealistic manner, enabling users to quickly and reliably prototype their creations.

assets/airenderings.jpg

Real-Time Live!

Here we show some live demos of Digital Salon. Certain parts of the operations have been shortened to make the video more concise. Note that these demos are previz focused on demonstrating their real-time performance. Better renderings can be achieved with more advanced offline physics-based renderers.

License

The video showcased contains renderings using Adobe in-house data, with all visual content subject to copyright regulations. The content may be freely used by the media with appropriate citation. Please credit Digital Salon and Adobe Research when using or referencing this content. Please contact yizho@adobe.com for commercial and non-commercial usage of the contents in the webpage.

BibTeX

@article{digitalsalon,
    title={Digital Salon: An AI and Physics-Driven Tool for 3D Hair Grooming and Simulation},
    author={He, Chengan and Herrera, Jorge Alejandro Amador and Zhou, Yi and Shu, Zhixin and Sun, Xin and Feng, Yao and Pirk, S{\"o}ren and Michels, Dominik L and Zhang, Meng and Wang, Tuanfeng Y and Rushmeier, Holly},
    url={https://digital-salon.github.io/},
    year={2024}
}