Meijie Hu · CLIENT: Meijie Hu
Persona AI is a software application for creating three-dimensional character faces. Users describe facial features, age, gender and personality in text, and the application generates photorealistic faces. Generative adversarial networks produce the facial textures, while three-dimensional morphable models define the facial geometry. Three variations are offered for each prompt, and single features can be refined through further text prompts without changing the rest. Finished faces can be shared by web link or exported at an adjustable resolution.
Press Kit
- Unique Properties
- Traditionally, crafting a photorealistic 3D character face requires hours of work using complex design platforms. Leveraging advanced AI models like GANs for high-quality facial texture maps and 3DMM for precise facial geometry, Persona.ai eliminates this complexity by allowing users to generate and edit photorealistic 3D character faces by simply describing the character’s facial features through text, making it accessible and effortless for 3D hobbyists and creators of all skill levels.
- Tags
- AI, UI/UX, creation tool, 3D design tool, user interface, human-centered design, 3D
- Production Technology
- Persona.ai leverages Generative Adversarial Networks (GANs) for high-quality facial texture synthesis and 3D Morphable Models (3DMM) for precise facial geometry, combining them to generate lifelike 3D character faces. For prototyping, I used Unreal Engine to generate character mockups, Blender to refine geometry, and Figma to design the interface.
- Design Challenge
- The biggest challenge was preserving creators' full creative control. Interviews revealed that artists often visualize a character before designing it in 3D and feel frustrated when generated results are too random to align with their vision. To address this, I redesigned the process to generate three variations instead of one, giving users more options to choose from. Additionally, I introduced secondary editing, allowing precise adjustments to facial features, ensuring greater accuracy and creative satisfaction.
- Project Duration
- The project started in Oct 2023 and finished in Feb 2024.
- Operation Flow
- Users start by entering a text prompt describing facial attributes, age, gender, and personality to generate high-quality facial textures and base geometries. Persona.ai then outputs three variations, and the user selects one for reinforcement learning. They can refine specific facial features using text prompts without altering the rest. Changes are saved in history. Once satisfied, users can share via a web link or export to their target platform with adjustable resolution and rigged body options.
- Research
- I conducted five interviews and surveyed 20 3D artists, game developers, and hobbyists to identify their character creation journey, pain points in their workflows, key character details they prioritize, and preferred file formats. After developing a prototype, I tested five concept cards on ways to use text prompts and text suggestions to optimize LLM character creation. I also A/B tested two final user flows with the original interviewees. Insights from the testing were used to refine the design and ensure users' full creative control.
- Inspiration
- As a 3D hobbyist, I know how frustrating it is to create realistic character faces without spending hours in complex software. Many fellow creators share this struggle, stuck between imagination and technical barriers. This inspired me to read AI-driven solutions in computer graphics research papers, leading to Persona.ai, a tool that turns simple text prompts into detailed 3D characters, making high-quality character creation easy and accessible for everyone.
Press releases are available in 1 languages.