Claude + HeyGen: a training-video field note

On a client project, I used HeyGen to generate a series of training videos. I came away impressed by the production quality, with one noticeable limitation.

These observations describe that project and the tools as I used them at the time.

The workflow

We started in Claude, grounding it in previous training videos so it could learn the tone, structure, and level of detail we wanted. From there, we developed the training scripts.

We then used HeyGen’s Video Agent to turn those scripts into finished videos.

The scripting process mattered. Giving the tools a clear sense of the material and presentation style made the production step more useful.

What stood out

HeyGen did a strong job with on-screen direction and presentation logic. It seemed to understand when a concept called for full-screen text, when an overlay made more sense, and how to pace visual emphasis against the script.

That part felt thoughtful. The body language and facial expressions were solid, and the voice sounded natural.

Where it fell short

Mouth-to-audio synchronization was the weak point. It was close, but not convincing enough to fully preserve the illusion.

For training content, though, the workflow was compelling: a strong scripting process upfront, followed by AI-assisted video production. The useful question is whether the quality fits the job the video needs to do—and which imperfections the audience will notice.