Character Designer built with Stability.ai and ChatGTP Turbo
April 8, 2024
Summary
As a Product Designer, I co-led the creation of Character Designer, a generative AI-based web app developed during the AI Encode Bootcamp. The tool empowers users to design unique characters by generating high-quality images and detailed narrative backstories using OpenAI’s GPT-3.5 Turbo and StabilityAI’s Stable Diffusion. My role included user research, ideation, prototyping, and usability testing, ensuring a gamified and user-friendly experience aligned with the needs of creatives.
Timeframe
3 weeks (2024)
Key Objectives
Develop a GenAI tool capable of generating images and backstories with user customization.
Design a user-centered gamified interface that aligns with the needs and user goals.
Ensure consistent visual quality: Integrate a stabilization feature for consistency across various character styles.
Validate the tool’s usability: conduct usability testing and gather feedback..
Results delivered
AI Integration and Scalability:
Generated over 100 unique characters, combining visuals and backstories.
Achieved a 95% user satisfaction rate, as measured through post-testing surveys.
Improved Usability:
Designed a gamified interface informed by insights from 8 creatives (via interviews and testing).
Reduced task completion time by 30% through 7 iterative design revisions, enhancing efficiency and satisfaction.
Rapid Development and Delivery:
Delivered a fully functional prototype within 21 days, demonstrating adaptability and scalability.
Industry Validation:
Received positive feedback from two senior AI experts, highlighting the innovative UI and seamless integration of generative AI technologies.
Concept and Experience
Character Designer was envisioned as a tool to assist creatives in the character creation process, helping them generate detailed visual and narrative content more efficiently. The app supports various art styles, customizable character traits, and user-driven storytelling, acknowledging that AI complements creative work rather than replacing it.
AI Model & Dataset Overview
Models Used: OpenAI GPT-3.5 Turbo for text generation and StabilityAI’s Stable Diffusion 1.6 for image generation.
Data: Utilized pre-trained models capable of handling diverse creative demands without additional fine-tuning.
Challenges: Fine-tuning the balance between image variety and consistency, ensuring outputs reflected user customization without diverging too drastically.
User Research and Design Process
Initial research included a questionnaire from 8+ creatives, revealing that users prioritized customizable traits and export flexibility. Given the niche nature of the target audience and the rapid 2-day timeframe, these insights were instrumental in shaping the design process. Based on these insights, I developed user personas and mapped user journeys that informed the design.
User Flow: Focused on intuitive navigation with clear steps for creating and saving characters.
Prototyping: Early sketches and wireframes were iterated upon using Figma, incorporating user feedback to refine design elements.
Prototyping and Iterative Design
Prototypes evolved through continuous validation with the team and potential users. The UI was designed to be playful and inviting, featuring a palette of Soft Peach (FEDADB), Lavender Mist (E4D9FF), and Mint Green (DDF5DA) for an accessible yet vibrant look. 7 iterations were created from user feedback and team validations.
Design System: Preline UI and Material Design components were used for faster implementation.
Feedback Loop: A usability testing workshop involving some of the users that previously answered the questionnaire led to further refinements, such as clearer navigation and streamlined interaction paths.
Development and Integration
The development phase focused on integrating APIs for GPT-3.5 and Stable Diffusion. Image stabilization techniques were implemented to maintain visual quality and consistency.
Challenges: Ensuring stable output while maintaining diverse customization options, running locally to not pay for ChatGTP.
Solutions: Adjusted API parameters for consistent results and optimized server-side processes to handle real-time requests, using ChatGTP.
Industry Feedback
Industry experts praised the intuitive UI and engaging demo. Constructive feedback included suggestions for integrating image generation within story creation for a more cohesive experience.
“It looks great overall; nice work! Cool idea, UI is very pleasing, the demo video is engaging.”
Matheus Pagani, AI Bootcamp instructor, Co-Founder of Venture Miner
“You could consider adding the ability to generate images within the story itself, featuring the characters. The app looks fantastic—congratulations on the great work!”