Summary
This full guide from Youri van Hofwegen explains how to build a realistic AI avatar that keeps the same face, voice and look across different scenes, using the Higgsfield platform. The core idea is to start from a single character sheet and reuse it for every later generation.
The process begins in Higgsfield’s image workspace with GPT Image 2 at 2K resolution and a 16:9 aspect ratio. A collage of reference photos produces a sheet with a full-body view and a close-up on a plain white background, with detailed skin texture such as pores, fine lines and freckles to avoid the smooth, plastic look common in generated faces. The sheet is then used to place the avatar in four different locations, including a prehistoric cave.
Next, the locations are animated in one 30-second, 1080p generation with Seedance, which handles the cuts between scenes itself. The video explains why dialogue should be written into the prompt rather than attached as an audio file, and why a script of around 75 words is needed to fill 30 seconds without repeated lines. The guide finishes with Higgsfield’s voice change tool so the avatar sounds like the creator.
📺 Source: Youri van Hofwegen · Published September 24, 2026
🏷️ Format: Tutorial Demo
