SundaySky · UX/UI Lead Designer

AI digital avatar

Personalize videos with AI avatars: fast, flexible, scalable.

AI feature Editor SaaS User testing Q2 2024
An AI avatar placed in a video scene with selection handles, the stock avatar library open beside it, and the caption “I'm Chris! Let me walk you through some helpful account features.”
16.2 % Account adoption
5.5 % Visitor adoption
73 % Moved or resized immediately

Pendo · post-launch

Goal

Enable users to add AI-generated avatars easily to their video scenes, helping them create engaging, human-like presenters without needing to film or hire talent, while maintaining full creative control.

The challenge

Many users avoided video presenters because of a lack of resources, production complexity, or discomfort recording themselves. Existing stock avatars felt generic and disconnected. The challenge was a scalable solution that felt personal, professional and easy to use.

My role
UX/UI Lead Designer
Team
PMs, Engineers, QA, Research, CS
Tools
Figma, FullStory, Pendo, Jira
Timeline
Q2 2024
Process
Discovery → Prototype → Design → Develop → Analyse

Discovery

2 inputs

01. Target user · marketer

Wants a human presenter at scale, without the cost of filming one.

Goal

Produce human-style videos quickly, personalize campaigns at scale, and maintain consistency by reusing avatars and voices.

Pain points

High production costs, limited avatar flexibility, and needing confidence that AI content won't hurt engagement.

Tools

  • Canva
  • Synthesia
  • HubSpot
  • Adobe suite
  • Loom
  • Google suite

Needs

Credit control, reusability, creative freedom.

Portrait of the marketer persona, recording at a microphone.
That's amazing. If I could get an AI picture of myself and just me talking, and not having to do it; that would be such a lifesaver.

User interviews · feedback sessions

02. Benchmark & inspiration

Market leaders emphasized realism, personalization and smooth workflows.

  • D-ID
  • HeyGen
  • Creatify
  • Powtoon

Competitive benchmarking of top avatar and video platforms across 2023–24. Most featured human-like avatars for business, with stock libraries and limited customization. Advanced tools emphasized personalized avatars and voices. Weaknesses included poor lip-sync and limited editing flexibility.

Competitor avatar platform: stock avatar library and scene composition. Competitor avatar platform: presenter placed over a branded background. Competitor avatar platform: voice and avatar pairing controls. Competitor avatar platform: customization options for a personalized avatar.

Internal benchmark review · 4 competitors

Prototype

2 rounds of validation

01. Interviews · concept validation

Clear interaction and strong interest, but concerns about realism and trust persist.

“It was really easy to figure out; I knew exactly what to do.”

Enterprise client

“I'd absolutely use a custom avatar if it looked like me.”

Marketer

We showed the initial avatar prototype to enterprise clients and marketers, collecting feedback on six aspects. The UI was praised as clear, intuitive and easy to navigate, and users valued switching avatars and voice options. Many were interested in ongoing use, especially with custom avatars. The limitations noted were avatar realism and lip-sync quality.

Concept validation results scored across six aspects of the avatar prototype.

Concept validation interviews · enterprise clients and marketers

02. Interviews · follow-up

Overlay avatars preferred, with strong demand for flexibility and control.

“I love the overlay: it gives us the freedom our clients expect.”

Enterprise marketing professional

“We'd use employee avatars for LinkedIn videos.”

Enterprise marketing professional

A live demo of the refined prototype with enterprise marketing professionals. Overlay avatars were preferred for flexibility, and demand for stock versus custom avatars varied by context. There was interest in multiple avatars per scene, and requests for hand movement, fade-in and voice-avatar sync.

Follow-up interview findings comparing overlay and full-scene avatar placement.

Follow-up interviews · live prototype demo

Design

5 features

The feature, working

An avatar chosen, placed and speaking inside a branded scene.

The avatar library open beside a finished scene reading “Thank you for choosing us! Hi, I'm Sabrina!” Watch the walkthrough

Final UX flow

From library to placed, previewed and voiced avatar.

End-to-end flow for adding an avatar: choose, place, preview, set voice, apply.

Design system

Built inside SundaySky's system, extended only where the feature required it.

  • Worked within SundaySky's design system.
  • Added new components for the avatar placeholder.
  • Added voice selection to the existing narration placeholder.
Design system sheet: components, buttons, notifications, icons, typography and colours.

Solution · main features

Feature 01. Add / replace avatar

Decision: an avatar library beside the media library

To align with the media library, we introduced an avatar library in the left panel with three options: custom avatars, with promotion for credit-based uploads; stock avatars; and “used in video” for quick re-use. That keeps the workflow consistent and makes avatars easier to discover and insert.

Avatar library in the left panel, with Custom, Stock and Used in video tabs.

Feature 02. Preview avatar

Decision: preview the way media already previews

We aligned the avatar preview with existing media previews and thumbnails. Users can play predefined text samples to check lip-sync and appearance, then adjust the voice later from the voice library: a familiar and flexible experience across all assets.

Avatar preview playing a sample line so lip-sync and appearance can be checked.

Feature 03. Scale, resize & align

Decision: knobs, handles and a scene toolbar

We introduced knobs, handles and a toolbar to move, resize and align avatars within the scene, giving users precise control to fit an avatar into any layout.

An avatar being resized in the scene using corner handles and the scene toolbar.

Feature 04. Avatar view

Decision: close-up and full-body modes

Two distinct view modes, close-up and full-body, so users can adjust how the avatar fits their scene. That flexibility allows better alignment with different layouts and storytelling needs.

Close-up and full-body avatar view modes compared in the scene.

Feature 05. Voice control

Decision: voice lives in the narration panel

A dropdown was added to the narration panel for voice selection whenever a video includes an avatar. This lets creators use the avatar's voice even in scenes without a visible avatar, keeping a consistent “presenter” across the whole video.

Voice selection dropdown inside the narration panel.

UX challenges

  • Creating a seamless flow for avatar insertion, preview and editing in a dynamic video canvas.
  • Balancing clear features with a minimal UI: guiding without constraining.
  • Enabling scalable personalization, including voice reuse and animation, without overwhelming new users.
  • Aligning with evolving AI infrastructure across voice, image, data and brief.
  • Supporting future custom avatars, including voice cloning and user uploads.

Analyse

4 measures

01. Adoption · Pendo

High initial interest, then a sharp drop.

16.2 % Account adoption
5.5 % Visitor adoption

Initial interest was high after launch, but usage dropped sharply, indicating that the feature delivered curiosity value but lacked enough quality and depth, such as avatar switching and personalization, to sustain ongoing engagement.

Pendo adoption data

02. Data path · Pendo

Entry worked; the steps after it did not.

73 % Immediately moved or resized

The entry flow worked well: adding and positioning avatars felt simple. Preview and apply steps were underused, and frequent replacements revealed quality and trust issues.

Pendo path analytics

03. Interaction patterns · Pendo

Curiosity was high, but trust in the output was not.

~35% Replacement rate
~40% Preview rate

Interaction patterns revealed strong initial curiosity. But frequent avatar replacement and limited use of preview actions suggested gaps in perceived quality and trust. This insight guided our focus on improving avatar realism, preview clarity and customization depth.

Pendo interaction analytics

04. Usability & workflow · FullStory

Most users interacted smoothly, but the thumbnail promised something it didn't do.

Session observations showed that users often expected the avatar thumbnail to trigger an immediate preview, but the interaction did not match that mental model. The overall flow was intuitive, though preview behaviour could be clarified to better align with expectations.

FullStory session observation

Insights and next steps

These improvements will make AI avatars more realistic, flexible and dependable: strengthening their role as the video's human touchpoint.

01

Engaging

AI avatars increase engagement and personalization potential across videos.

02

Preview flow

Clarify the preview experience and make feedback more visible.

03

Customization

Add more voices, avatars and style variations for greater creative control.

04

Quality

Enhance realism through better lip-sync, motion and facial expression.

Template applied to your heaviest page. 18 images, the walkthrough video and five real figures, all from your existing page. The hero uses your transparent avatar composite rather than a framed screenshot; the video only loads from YouTube once you click play; figures count up as they come into view.