This project aims to enhance user creativity in collage creation through the integration of AI technology, while simultaneously accumulating the team's experience in AI feature development.
In a series of generative AI projects, I collaborated with two iOS engineers on the Magic Avatar and AI Artspace Page projects, responsible for the UX and UI design process and deliverables. For the AI Photo Album, I participated in the UI and visual design.
During the generative AI project, there were numerous technical limitations. It was necessary to coordinate with the iOS and back-end development teams, as well as the product team, and to launch the first version of the MVP within a limited timeframe.
The generative AI project spanned nearly 3 months of development and underwent algorithmic iterations. The generative AI functionality contributed to a 0.3% increase in the product's overall daily revenue. For the Magic Avatar feature, one month after launch, the daily revenue was $133 and the monthly revenue reached $4000. Revenue and retention gradually decreased after the initial surge.
Since 2022, the rise of AI tools like Stable Diffusion, Midjourney, and ChatGPT has revolutionized the image software industry, leading to increased investment in Generative AI tools. This has notably influenced the image collage sector, with products such as Lensa, Meitu, Snow, and EPIK introducing AI avatar features. Snow expanded this with AI-generated profile and ID photos, adapting AI avatars from novelty to practical use.
To enhance our product's AI capabilities, we initially focused on the Magic Avatar feature, followed by the development of AI Photo Album features.
Highlighting our year's progress, I curated an AI feature page "AI Artspace", offering users a comprehensive experience of our generative AI features.
This project aims to enhance user creativity in collage creation through the integration of AI technology, while simultaneously accumulating the team's experience in AI feature development.
1. Limited Time
The goal for Magic Avatar was to launch a version of the MVP within one month, with numerous tasks to be completed. Therefore, it was essential to break down the tasks according to the main process and implementation details and prioritize them to deliver within the deadline.
2. Techical Limitations
The Taipei team lacked algorithm-related talent, necessitating reliance on Beijing's R&D and external resources to support our AI feature development. We planned to switch to our self-developed model once the service stabilized.
3. Generating Explicit Images
We adjusted negative prompts to avoid generating NSFW images.
During February to March 2023, the project capitalized on the Web3 and AI technology trend, along with the Y2K revival and Gen Z's preference for self-expression, to launch Magic Avatars. The initiative aimed at both gaining AI development experience and increasing revenue by attracting new users interested in self-expression through consumable purchases.
This project underwent 2 to 3 major iterations, focusing initially on the main process (purchase) and style selection, and later adding a repeat usage process for the same set of photos and promotional strategies for existing and unsubscribed users to boost business conversion.
This project did not include user interviews, so before diving deeper, we needed to understand why users would choose this service and the scenarios for using AI-generated profile.
The popularity of AI-generated profile in 2023 follows the 2022 launch of Midjourney and advancements in AI technology. With just a few simple keyword inputs, images can be generated, and generative AI products can create diverse avatars from just a few user selfies, significantly lowering the barrier to obtaining sophisticated avatars. Each unique AI generated profile highlights individuality, meeting the need for updated profile and easy sharing on social media, allowing everyone to show off to friends and family. Everyone can curating their ideal online persona through AI generated profile pictures.
Developing Magic Avatar not only accumulates experience in AI feature development but also provides users with various digital avatars for social sharing, enhancing user engagement. As a consumable service, it creates new revenue streams for the company. Furthermore, by analyzing user preferences, we aim to introduce more trending avatar styles.
Looking at Generative AI products, their visual style and metaverse products are similar, mostly adopting a Dark Mode theme, emphasizing a cool, black aesthetic and individuality. The primary audience includes tech enthusiasts and early adopters who are open to trying new products, so the design is tailored to their preferences. Furthermore, Dark Mode is used as the main visual theme for the AI feature to differentiate it from the app's main features (homepage, draft box, personal page, etc.), also creating a more immersive user experience.
Therefore, black, complemented by purple and gradients, is the base color scheme, with purple as the primary color for CTAs. Purple is chosen for its sense of mystery and is often used in contemporary pop culture to represent fantasy, magic, and creativity (Adobe).
Steps
We first let users see the available options for different genders and preview the styles before they decide to purchase. Therefore, we put a paywall after the upload photo step. Only after users have paid will their photos be officially uploaded to the server. This prevents users from seeing the generated results before they pay, which can reduce refunds and waste server resources.
.png)
In creating Magic Avatars, our service works by sending photos to a server via an API during upload. While the server was unstable, we considered a refund mechanism for any potential failures to safeguard user rights.
Failures might occur during photo upload. If users exit the app, it's treated as a restart with their purchased plans and points preserved. Another scenario is receiving an error message from the Beijing server. For example, if a task takes 1 hour, and users buy the same or different plans without knowing the outcome, success or failure isn't immediately clear. Points are initially deducted during server processing and are refunded if the task eventually fails (after server completion).
To facilitate easy tracking and verification of remaining points by the program and users, we use a notation of n*50 + m*100 + q*200 for remaining points, where 50, 100, and 200 represent the number of units sold to users.
.png)
.png)
If users have already created Magic Avatars, they can view their previously made packs on the generation page. By clicking 'Create new avatars', a 'Select Photo Group' sheet will appear, allowing users to use past photos or upload 10 new ones to generate new avatars. This feature makes it convenient for users who don’t want to upload a selfie again to produce different Magic Avatars.
.png)
💰 It resulted in an average of 6 members renewing daily, demonstrating that offering AI Avatars as a gift was an effective renewal incentive.
.png)
.png)
The generative AI project spanned nearly 3 months of development, undergoing algorithm iterations. The generative AI features contributed to a 0.3% increase in the product's overall daily revenue. However, it required significant resources and could not achieve breakeven in the short term.
One month after Magic Avatar's launch, its daily revenue was $133, and the monthly revenue reached $4000. However, revenue and retention gradually decreased after the boom.
Data indicates that the next-day retention for Magic Avatar is very low, with most sales concentrated on the $3.99 plan. This suggests that users are trying out the feature out of curiosity. To capitalize on the existing momentum of Magic Avatar, it is recommended to observe competitors, such as the popular EPIK AI Yearbook from 2023, keep up with social media trends, or expand the application scenarios of virtual avatars. For instance, creating professional photos for various industries that simulate the effect of a professional studio shoot, saving the high costs of makeup and hair, allowing users to apply them to their personal branding and social networks.
.png)
The R&D team spent a lot of time and cost on cross-team coordination and communication on AI technology and connectivity. The algorithm team is also constantly revising the model. Since the models were provided by the Beijing team, we relied on the designers and QA to test whether the avatars were undressed and to adjust the aesthetics of the generated avatars to match the public's aesthetics.
Furthermore, this project marked the company's inaugural foray into the development of AI features, entailing a significant investment in manpower. The short-term returns are not sufficient to offset the labor costs. It is recommended that we take a long-term view to monitor subsequent earnings and social media trends, continuously updating the models and enhancing server stability.
In this project, my role encompassed UI/UX and graphic design, while dynamic effects were skillfully assisted by Rong Wu. The opportunity to develop a feature distinct from PhotoGrid's main theme was particularly exciting. It allowed me to explore unconventional visual elements and provide users with a novel AI-powered experience. This project was a testament to the strength of collaborative creativity and proved to be a valuable platform for me to harness and amplify my design skills.
In response to a series of AI feature launches, I was responsible for the visual output of Magic Avatar, AI Album, AI Animation, AI Creativity, and AI Artspace.
Highlighting our year's progress, I curated an AI feature page "AI Artspace", offering users a comprehensive experience of our generative AI features.
The UI below, from left to right, consists of the subscription page and introduction popup for AI Photo Album, and the payment popups for AI Animation and AI Creativity.
In a nutshell, creating a multitude of visuals for AI features has been liberating for me, moving away from the constraints of tool-based UI design. It allowed me to boldly experiment with different visual styles and color schemes, expanding the boundaries of my design.
Shout out to the team; without them, these challenging projects would not have been possible.
Product Design - Wen Lin
Motion Design - Rong Wu
Design Director - Jesse Chou
iOS Team - Allen Hsu, Fong Lin, Kyle Peng, Vernty Chang, Enola Zhu
Backend Engineer - Moby Lin
Product Team - Taipei & Beijing Team