Writing the Script With One Prompt

Before you generate a single frame, you need to decide what the channel is even about — and that decision has money attached to it. YouTube pays more for some niches than others through RPM (revenue per thousand views), and the three highest-paying categories are finance, tech, and educational content. Pick your topic from inside one of those three, and you're already stacked in your favor.

For this walkthrough, the education niche was chosen — the same lane as Brightside, the channel used as the reference. Educational content tends to appeal to a broad audience and holds retention well, and retention is what YouTube's algorithm rewards with monetization.

With the niche locked in, the actual scriptwriting is handed off entirely. Claude Fable 5 (Anthropic's newest model at the time of recording) is paired with Higgsfield MCP, which connects that model to a full generation engine — image, video, and voice — in one system. Instead of researching topics or writing anything by hand, a single instruction is typed into Claude: analyze the target channel's scenarios and hooks, then write a script for a similar video, with a link to the Brightside channel pasted in alongside it.

From there, Claude does the reverse-engineering itself — studying the channel's most-viral videos, its hooks, and its structure, spotting the patterns that make them work, choosing a topic on its own, and returning a full, ready-to-use script. In this run, it picked the eruption of Mount Vesuvius as the topic and structured the script into nine beats, each mapped to a time range and a set of clips — from the opening hook, through the calm-before-the-storm setup, the eruption itself, the town being buried, the decision to flee or stay, the pyroclastic surge, the aftermath, and finally the archaeological discovery and closing twist. That beat-by-beat structure is what gives the finished video its pacing, and it's built to hook viewers from the very first second — which is the whole point of front-loading research into an AI model instead of doing it manually.

With the script finished, the next step is turning that structure into an actual video — visuals, consistent characters, and voiceover — using a single prompt to Higgsfield MCP.