How we keep videos accurate
Why AI video generation models can invent a number, a label or a result, and how Explain My Charity keeps a video true to your document: a script you approve, and every frame built in code from it.
Transcript
Ask an AI video model for a chart from your report, and it paints something that looks like one: from patterns, not from your data. To it, letters are shapes, so words come out garbled; a number only has to look right; and nothing ties its chart to your report. Independent tests, and their makers' own reports, agree. When we tried one with prompts from one of our own videos, it misspelt a name and a web address, and its chart had none of the real data.
What we do instead: our specialised AI agents draft the script, and you edit and approve it. Then our software draws every frame: typeset words, charts plotted from your report's own figures, diagrams in code. A synthetic narrator reads it. So every word, number and chart comes from your report and the script you approve.
A whole seventy-second video, drafting included, used about fifty to ninety watt-hours when we measured it in September. One take of AI-generated footage used one point seven kilowatt-hours. Try it yourself: get started today at explainmycharity.com.