video1.0<iframe src="https://www.loom.com/embed/58d5844542754ee2bdfd734e6e3bc6a7" frameborder="0" width="1920" height="1440" webkitallowfullscreen mozallowfullscreen allowfullscreen></iframe>14401920Loomhttps://www.loom.com14401920https://cdn.loom.com/sessions/thumbnails/58d5844542754ee2bdfd734e6e3bc6a7-ef534846c67b923b.gif528.256This Loom explains a step by step workflow for creating realistic AI avatar videos using motion reference and the Higgsfield prompt tool. The creator first records a steady, fixed-camera motion video and uses it as the motion reference, then selects or generates a character image, noting that front-facing shots and simple textures work best. Next, they recommend using Higgsfield’s Edit Video with the SeaDance model, building a strict hyperrealism prompt to replace the person while keeping clothing and other elements unchanged. Finally, they advise running the result through CapCut before posting to remove metadata, especially to avoid TikTok labeling it as AI generated.