HeyGen launches two-host AI video podcasts from documents and links
The workflow writes dialogue, assigns AI hosts, adds captions and B-roll, and lets users edit each line before rendering.
By Ryan Merket · Published
Why it matters
HeyGen is pushing AI video into repeatable marketing, training and internal-communications work, where script control and review matter as much as generation speed.
HeyGen, led by co-founders Joshua Xu (@joshua_xu_) and Wayne Liang, launched Video Podcast on July 29th, turning a topic, URL or PDF into an edited show featuring two synthetic hosts. The output includes studio scenes, captions, B-roll and camera cuts, giving users a finished video file instead of an audio conversation or an unedited avatar clip.
https://x.com/heygen/status/2082510248489943367?s=46
The Video Podcast product generates dialogue, assigns speaker roles and synchronizes each host's voice with lip movements. Users can revise individual lines, change speaker assignments, adjust tone and insert supporting facts before rendering the episode. HeyGen says the system can produce a 16:9 video in minutes, with longer episode limits tied to the user's subscription. (heygen.com)
The release extends the founding thesis Xu has pursued since 2020. Xu studied robotics and computer vision at Carnegie Mellon University before working at Snap, while Liang studied human-computer interaction at CMU. Xu told CMU's Masters of Robotic Systems Development program that his discomfort recording himself helped produce HeyGen's original goal: "replace the camera." The co-founders met through CMU, with Xu concentrating on engineering and Liang on design. (labs.ri.cmu.edu)
HeyGen is packaging the entire show
Video Podcast moves HeyGen beyond generating a presenter reading a supplied script. HeyGen is treating the episode as the unit of production: the system interprets source material, writes the exchange, stages two hosts, selects cuts and places supporting visuals around the conversation.
That packaging is the commercial point. A marketing or training team can feed HeyGen a report, product page or internal document and receive an asset formatted for YouTube, a company website or an internal content library. The same source can be recast as a branded discussion without scheduling presenters, recording footage or handing the result to a separate video editor.
The editor remains central because the system generates its own dialogue. HeyGen's launch thread emphasizes line-level review, voice previews and pre-render validation. Those controls give users a chance to correct terminology and factual errors before synthetic hosts deliver the material with the visual authority of a studio production.
Google's NotebookLM already moved into video
HeyGen's launch post says, "Everyone else stops at audio." That line is inaccurate as a market-wide claim. Google introduced NotebookLM's two-host Audio Overviews in September 2024, added Video Overviews in 2025 and released Cinematic Video Overviews on March 4th, 2026. The cinematic version uses Gemini, Nano Banana Pro and Veo to build animated visual narratives from supplied sources. (blog.google)
HeyGen is pursuing a narrower production format. NotebookLM is designed to help users understand source material and can generate several study and presentation artifacts. HeyGen is selling a publishable, presenter-led show, with visible hosts, recurring studio styles and editing controls aimed at marketing, education, learning and development, and internal communications.
That distinction also puts HeyGen in competition with avatar-video platforms such as Synthesia and DeepBrain AI, while general video generators including Runway, Adobe Firefly and OpenAI's Sora address a broader range of visual production. HeyGen's advantage will depend on whether the two-host format looks consistent through a full episode and whether teams can trust the generated script enough to make the workflow repeatable.
HeyGen is using its growth to widen the workflow
Xu said on June 25th that HeyGen had passed $200 million in annual recurring revenue, doubling in eight months. HeyGen also claimed over 30 million users, adoption across 85% of the Fortune 100 and 118 million videos created. Those figures are company-reported and do not break out how much revenue comes from individual creators, enterprise contracts or newer applications such as Video Podcast. (heygen.com)
HeyGen raised a $60 million Series A led by Benchmark in June 2024, with Thrive Capital, BOND, Conviction and SV Angel among the participating investors. Forbes reported that the round valued HeyGen at roughly $500 million. (forbes.com)
The current pricing structure gives HeyGen several routes to sell the new workflow. The Creator plan costs $29 per month, Pro starts at $49 per month, and Business costs $149 per month plus $20 for each additional seat. HeyGen says free accounts can test shorter Video Podcast outputs, while paid tiers provide longer videos and larger credit allocations. (help.heygen.com)
Video Podcast puts Xu's camera-replacement idea into a format companies already know how to distribute. The product's success will turn on the less visible parts of the pipeline: source interpretation, dialogue quality and the amount of human review required before an episode is safe to publish.