Two-host AI video podcast: what Sume avatar videos can do

A single Sume avatar video has one resolved avatar, so two hosts sharing one frame is not something the docs show. For a two-host format, render each host as its own avatar video and cut between them on a Timeline. HeyGen's July 2026 Video Podcast is a single product feature with two hosts in one scene.

What does HeyGen's Video Podcast do?

HeyGen says it "turns any script, URL, PDF, doc, or topic into a two-host video podcast in minutes", and that "Your two AI hosts share a real studio scene and react to each other". Both quotes are from its July 2026 release notes.

What does a Sume avatar video support?

The avatar video docs say current execution supports "one resolved avatar per final video" and expects scene backgrounds to resolve to one shared scene. Multi-scene video_inputs are ordered scenes of one avatar with voice or silence beats, not two people talking.

How do I fake a conversation?

Write the dialogue as alternating lines, render each host's lines as separate avatar videos with different avatar_handle values, and use the same scene prompt so the backgrounds match. Then sequence the clips on Timeline 1.0. The result is cuts between hosts, not a two-shot where one reacts while the other speaks; the earlier post on two-speaker avatar videos covers the same idea.

What are the costs of that approach?

Each clip is its own job with its own reservation, and each must be 4-60 estimated seconds. A 10-minute show cut into 30-second turns is 20 requests. Keep the same aspect_ratio across all of them.