Descriptions:
Creator Dan Kieft walks through a complete, reproducible pipeline for building consistent AI influencer videos for UGC-style product ads, tackling the two hardest problems in AI-generated influencer content: character consistency across scenes and voice consistency across clips.
The workflow combines NanoBanana 2 for initial character image generation (with detailed JSON-style prompting via Claude to produce more realistic, animatable faces), Sora 2 as the primary A-roll generator for the influencer speaking to camera, and Cling for B-roll footage. A key update covered is Sora 2’s new 20-second video capability, demonstrated with a Liquid Death-style product ad generated from a single image reference and a single prompt. For voice consistency — historically the biggest giveaway in AI influencer content — Kieft layers three fixes: audio prompting within Sora 2 (specifying accent, pitch, and delivery style), Adobe Podcast’s speech enhancement tool to strip noise and artifacts, and ElevenLabs’ voice changer to normalize tone across generations.
The video also shows how to use Claude with a custom markdown prompt template to generate detailed, specific character descriptions that produce significantly better NanoBanana outputs compared to generic prompts. For brands, agencies, or solo creators looking to produce scalable influencer-style ad content without hiring talent, this tutorial offers a concrete, tool-specific workflow with honest coverage of each tool’s current limitations.
📺 Source: Dan Kieft · Published March 19, 2026
🏷️ Format: Hands On Build
