DISCOVER
Start with our strongest curated characters.
03PRODUCT INSIGHT
WHAT MADE THE CHARACTER FEEL ALIVE
A believable AI characterneeded more thana good response.
Conversation, voice, expression, and visual context had to respond together—consistently and in real time—for the character to feel truly alive.
CONVERSATION · EXPRESSION · PRESENCE
04CHALLENGE 01 · VOICE CONVERSATION UX
TURN-TAKING CLARITY
Voice made the AI feel alive. State clarity made it usable.
A live exchange can become ambiguous quickly. The interface had to make speaking, listening, waiting, and response timing legible without interrupting the conversation.
- 01Problem
Voice, text, and visual response competed for attention.
- 02Principle
One dominant state at a time; state clarity before decoration.
- 03Outcome
Speaking, listening, waiting, and responding became distinct states.
05CHALLENGE 02 · CREATE YOUR OWN MEL
THREE WAYS TO BEGIN
Creating your own MELneeded more than oneway to begin.
Some users wanted a polished character immediately. Others had a specific face in mind, while some wanted control over traits such as age, personality, and skin tone. I designed three creation routes—Discover, Photo, and Custom—so users could begin with the level of intent and effort that felt natural to them.
- 01Problem
A single creation flow assumed every user already knew exactly what they wanted.
- 02Principle
Match the starting route to the user’s level of intent.
- 03Outcome
Three routes matched different levels of creative intent and control.
PHOTO
Create a character from a visual reference.
CUSTOM
Define the character through individual traits.
INSPIRATIONREFERENCECONTROL
THE MOST NATURAL STARTING POINT
Users preferred reactingto a complete characterbefore defining one.
Discover became the most frequently chosen creation route. Curated presets required the least upfront effort and gave users confidence in the final quality before they had to define individual traits.
WHY AN OVAL WHEEL?
Extending the brand languagefrom circles into interaction.
MEL’s visual identity was built around circular forms. I extended that language into a vertical oval to better frame portrait-oriented character imagery within a mobile screen. The centered character remained visually dominant, while partially visible characters at both edges signaled that more options could be explored.
Compared with a grid, the wheel made rapid comparison less efficient but gave each character more visual presence and made browsing feel closer to meeting characters one at a time.
The wheel changed browsing from comparison to discovery.
06CHALLENGE 03 · GENERATIVE VISUAL PRODUCTION
FROM POSSIBILITY TO PRODUCTION
Generated visuals had to be more than technically possible. They had to be product-ready.
At the time, generative models could produce striking single images, but character identity, anatomy, composition, and scene continuity were still unreliable.
I worked closely with the ML team to turn visual judgment into a repeatable ComfyUI-based production workflow. I defined the emotional brief, scene hierarchy, composition, continuity, and product constraints, then reviewed and refined the outputs until they could function as real date and location scenes—not just technical demonstrations.
The technology showed what was possible. Our workflow made it usable.
- 01 · DEFINETranslate each relationship moment into a visual brief.
Mood, relationship context, location, camera direction, and product constraints.
- 02 · BUILDCreate repeatable generation conditions with the ML team.
ComfyUI nodes, prompts, references, seeds, and controllable inputs.
- 03 · EVALUATEReview every output through a product-quality lens.
Character identity, anatomy, composition, artifacts, scenario clarity, and UI legibility.
- 04 · SHIPAdapt selected visuals for real product scenarios.
Date scenes, locations, crops, safe areas, and platform-specific formats.
07IMPACT & REFLECTION
REPORTED BETA SIGNALS
A more coherent AI experienceshowed stronger engagement signals.
Reported average chat duration
Reported onboarding exit rate
Beta participants
Metrics are retained from the original beta materials and presented as reported directional signals rather than causal proof.
A believable AI companion is not created by model capability alone. It emerges when conversation, voice, character creation, and generated scenes work together as one understandable experience—giving users a reason to stay and continue the relationship.
What this experience opened next
Next project · 03 / 04 KKuuk Tactile interaction · Haptics · 2026From guidance to sensory feedback
After reducing hesitation in an AI creation flow, I explored a different question: can immediate sound, motion, and haptics make interaction itself worth repeating?