How we test

Status: hands-on testing started on 21 Oct 2026. Opus Clip, Vizard and Submagic are tested (results), and so is Pictory (results), which we also ran on our own sample lesson script. The avatar tools Synthesia and HeyGen were run on free plans with an excerpt of that script (results). Descript’s editor was run on the free plan with the full sample lesson, and we measured transcript accuracy against our script (results). Riverside is not tested: its pricing page showed no free plan and its trial asked for a card. Every other tool is labelled “Not yet tested”.

1. Same inputs every time. Two test files: a 3:46 sample lesson that LessonBench made and owns (slides plus narration, with captions and a plain script for avatar and caption tools), and the first 45 minutes of a public-domain university lecture (720p, one lecturer on camera, with slides) for clip makers. Lecture footage: “The Federal Reserve and the Financial Crisis, Part 1” (2012), Board of Governors of the Federal Reserve System. Public domain. Not affiliated with or endorsed by the Federal Reserve.
2. Same settings. Free plan or a free trial that doesn’t need a card, default settings, 9:16 vertical output. We note anything we had to change by hand.
3. Recorded results. Processing time, number and length of clips, spot checks of clip starts and ends and of captions (we say how many we checked), speaker framing, watermark and export limits on the plan tested, and dated screenshots. For editors with transcripts we also measure word error rate against our own script. Filler removal on real speech and horizontal export are not part of the first round.
4. Plan and price checked on the date shown. We link the vendor pricing page rather than guessing.
5. No fake hands-on. Untested tools are labelled “Not yet tested, based on vendor documentation”.
6. Who tests. An AI-assisted team runs and documents each test and keeps dated screenshots.

We plan to retest major tools roughly every quarter and note the update date.