paint-brush
Solving the Uncanny Valley Effect with Meta Avatars, Using AIby@gabrelyanov
239 reads

Solving the Uncanny Valley Effect with Meta Avatars, Using AI

by ASHOT GABRELYANOVOctober 20th, 2023
Read on Terminal Reader
Read this story w/o Javascript
tldt arrow

Too Long; Didn't Read

In 2021, during Meta's grand unveiling of its metaverse vision, the company introduced avatars as the designated digital identities for self-expression in the virtual world. However, this initiative faced backlash, with some industry experts labeling the avatars "Cringe". Our team delved into this matter, seeking insights from professionals to understand why Meta Avatars fall short in aesthetic appeal.

People Mentioned

Mention Thumbnail
featured image - Solving the Uncanny Valley Effect with Meta Avatars, Using AI
ASHOT GABRELYANOV HackerNoon profile picture

In 2021, during Meta's (formerly Facebook) grand unveiling of its metaverse vision, the company introduced avatars as the designated digital identities for self-expression in the virtual world. However, this initiative faced backlash, with some industry experts labeling the avatars as "Cringe."


Meta Avatars. Source: Meta


Our team delved into this matter, seeking insights from professionals to understand why Meta Avatars fall short in aesthetic appeal and explored an alternative AI-based technology.

Reasons for the Lack of Appeal in Meta Avatars

  1. Uncanny Valley Challenge: Meta's avatars face difficulty in finding the right blend between a cartoonish and humanoid look. This results in the well-known "uncanny valley effect," where people experience discomfort when a humanoid avatar almost, but not entirely, resembles a real human. This sense of unease undermines the effectiveness of these avatars.


    Uncanny Valley Effect Visualization. Source: Mac Reddin, LinkedIn


  2. Overly Radiant Skin Tone: A notable concern arises in the skin shaders of Meta's avatars, creating an excessively glossy, plastic-like texture. This unrealistic shine takes away from the avatars' charm and authenticity, diminishing their relatability.


3. Flawed Eye Design: Another evident issue lies in the design of the avatars' eyes. Specifically, the placement of pupils is inaccurately close to the lower eyelid. This departure from natural eye anatomy enhances the unsettling nature of Meta avatars, introducing epicanthus to everyone, irrespective of its natural presence.


However, as we explored avatars, we found another problem in other avatar technologies that we had to solve too.

Laborious Avatar Creation in Alternative Apps

Apple's avatars and Snapchat's Bitmoji, widely used in messaging, employ a manual avatar creation toolkit. Users must meticulously design their digital twin by choosing from a wide range of facial features, such as eyebrows, chins, and noses.


Manual Avatar Creation in Apple’s Memoji and Snap’s Bitmoji


This method has three main drawbacks:

  1. Creating a realistic avatar requires users to possess specific artistic skills.

  2. The user experience can be daunting and complex.

  3. It imposes a considerable workload on developers, as they need to generate numerous versions of distinct facial components to meet varied user preferences.

    Our Avatar Maker's Machine Learning Approach to Avatar Generation

    Hence, our decision was to develop an Avatar Maker that achieves the balance between a cartoony and realistic appearance, steering clear of the uncanny valley effect. Moreover, it enables users to effortlessly craft their avatars using just a selfie photo, eliminating the need for laborious manual construction.


    AI-powered Avatar Generation Tech Demo of Avatar Maker


    Importantly, this technology functions locally on mobile devices, safeguarding users' sensitive data, especially facial images, by avoiding cloud-based processing.

    Here's a breakdown of how it works:


    The process of generating avatars begins with the creation of a 3D face model using a single selfie. A neural network utilizes this photo to generate a 3D mesh representation of the user's head.


    3D Mesh Extraction from 2D Selfie


Following this, the mesh undergoes an automated stylization process, giving it a more cartoonish appearance, serving as the foundation for the avatar's head. 3D Mesh Stylization

Subsequently, additional neural networks come into play. They classify and synthesize various facial features like hair, eyebrows, facial hair, and color, adapting them to the user's selfie-based style. Facial Features Detection for Avatar’s Personalization

In the case of users wearing glasses, a specialized neural network identifies and generates a 3D version of them.


Eye Glasses Auto-generation


Additionally, users have the option to apply various customizations such as headwear, makeup, bindis, and more to infuse their avatars with a vibrant and funky style.


Avatar Maker iOS App Customization


We accomplished this by gathering a distinctive collection of 20,000 3D High Fidelity Face Scans, showcasing a variety of individuals in terms of races, genders, and ages. This exclusive compilation plays a pivotal role in empowering Avatar Maker's customization and elevated facial performance features.

Face Scans Dataset Collection

How to Try Our AI Avatar Maker

Initially launched as an MVP on iOS, Avatar Maker is now accessible as a mobile app. It enables users to effortlessly create personalized avatars and craft animated sticker packs for WhatsApp or GIFs. Explore Avatar Maker today and share your creative results!


Get Avatar Maker for iPhone. Try Avatar Maker today and share your results!


Avatar Maker