Descriptions:
Dan Kieft walks through the full pipeline for creating a realistic AI avatar clone using Kling’s Cance platform, accessed through Heyfield, arguing that the technology has improved dramatically compared to tools available nine months earlier. The video opens with a side-by-side demonstration of his AI-generated avatar versus his real self, highlighting improvements in skin texture detail, natural dialogue pacing, and ambient light rendering that make the output difficult to distinguish from real footage.
Kieft covers two approaches to voice input: recording directly with Audacity (free, but Cance will alter the voice somewhat — expressive inputs produce better results) and using ElevenLabs professional voice cloning for higher fidelity. He notes that Cance audio references must be under 13 seconds and explains the tradeoffs between the two methods with audio comparisons. For video generation, he introduces “timeline prompting” — a structured format that breaks the scene into explicit temporal segments — arguing that specificity in the prompt is the primary driver of realism.
The talk also includes a comparison between Cance and HeyGen, with Kieft favoring Cance for its reproduction of natural pauses, eye movement, and head behavior. Practical use cases discussed include replacing studio recording sessions, scaling content production, and building public-facing video avatars. Viewers should note that Cance is positioned here as a paid feature within the Heyfield platform.
📺 Source: Dan Kieft · Published May 19, 2026
🏷️ Format: Tutorial Demo
