Yeah AI avatars—the user interface of AI voice chat—come in two distinct forms: Video AI Avatars and 3D Modeled AI Avatars.

Video AI Avatars

Video avatars are much easier to create but are limited in their range of body actions, typically being just a torso or a “talking head.” Despite these limitations, they often feel more lifelike, as they use actual video footage of a person talking in response to your questions. However, unless the footage is of very high quality, the resolution and realism of video avatars are constrained by the original video recording. Additionally, making video avatars of historical figures or deceased individuals typically results in lower definition and quality due to the limitations of available source material.

The process of creating a video avatar generally involves recording a 2- to 5-minute video of yourself, which should include moments of speaking, listening, and silence. Lighting should be bright, and the background should not be busy. Depending on the platform, some require high-definition audio, while others specifically recommend avoiding it. These platforms are evolving rapidly, especially since this technology only became widely available at the end of 2024.

Once the video is recorded, it is processed to generate the avatar. It’s important to note that there is no post-processing or editing of the avatar’s appearance after generation. Several platforms currently support the creation and streaming of video avatars, but they lack the ability to have a custom RAG knowledge base, which is our premium solution.

3D Modeled AI Avatars

3D modeled avatars, like those used in movies and video games, are much more complex and time-consuming to develop but offer full, realistic body movement. Unless you’re working with a top-tier, high-end model, 3D avatars may still appear animated when compared to video avatars, which are sometimes jumpy.  The technology is evolving.

The creation of a 3D modeled avatar typically starts with a full-body blockout to define proportions and general shape, including placeholder clothing. The model is then refined through detailed sculpting, followed by the addition of realistic hair using hair cards or similar techniques. Surface details and textures are added to replicate skin tone and material appearance.

Rigging provides the model with a skeleton for movement, and weight painting ensures smooth deformations during animation. For highly realistic avatars, photogrammetry—the process of generating a base model from a series of photographs taken in a specialized studio with hundreds of cameras—can be used. This significantly reduces sculpting time while increasing detail and realism.

3D modeled avatars are particularly well-suited for applications like holographic museum displays, where realistic movement and full-body interaction are essential. 

The images below provide examples of the difference in appearance between AI video AI avatars vs 3D modeled AI avatars.  Note that the more effort put into the development of the 3D modeled avatars results in a more lifelike image.  At least with these examples, the 3D avatar looks to be a little more on the crazy side.