HeyGen moves beyond talking heads
HeyGen has spent years building its reputation on avatar and talking-head tools: you write a script, pick an avatar, and the platform animates a face reading your words. On September 30, 2026, that changed. HeyGen launched HeyGen Video, its first general-purpose text-to-video model, which generates an entire scene from a single prompt, subject, setting, lighting, and sound, in one call.
That's a meaningfully different product from Avatar V. Avatar V animates a look you already own from a script you already wrote. HeyGen Video invents the subject, setting, light, and sound from a description. For creators, that means it's less a presenter tool and more a scene generator, closer to what you'd use for b-roll, cutaways, or stylized establishing shots than for a talking-head explainer.
What it actually generates
Each clip runs 5 to 15 seconds and comes with its own audio track, dialogue, ambience, and sound effects, generated together rather than bolted on afterward. HeyGen says there's no separate speech pass, no lip-sync step, and no avatar to select. You describe the scene, and the model produces picture and sound as one unit.
The model is built on MiniMax H3 and post-trained by HeyGen, and it supports a range of styles, from live action to stop-motion. Resolution options span 480p, 768p, 1080p, and 2K, with 1080p and 2K billing at 3x the 768p rate. Aspect ratios cover the usual spread: 21:9, 16:9, 4:3, 1:1, 3:4, and 9:16. Output is MP4 (H.264) at 24 fps with AAC stereo audio, which should drop into most editing pipelines without conversion headaches.
- Clip length: 5 to 15 seconds, generated with synced audio
- Resolutions: 480p, 768p, 1080p, 2K (1080p/2K cost 3x the 768p rate)
- Aspect ratios: 21:9, 16:9, 4:3, 1:1, 3:4, 9:16
- Output: MP4 (H.264), 24 fps, AAC stereo audio
- Access: HeyGen API and partner platforms
Pricing and the reproducibility angle
HeyGen Video is priced from $0.01 per second through October, available via the HeyGen API and partner platforms. For creators testing ideas before committing to a final render, HeyGen points to a 480p tier at $0.02 per second as a way to lock a shot cheaply before rendering it once at full quality.
HeyGen also emphasizes reproducibility: fixed seeds return the identical clip for the same prompt. Practically, that means refining a shot is a matter of changing one clause in your prompt rather than re-rolling the whole generation and hoping for something closer to what you wanted. If that holds up in practice, it's a real workflow advantage over models where every generation is a fresh gamble.
The leaderboard claim, and the caveat that came first
On October 22 to 23, 2026, HeyGen announced that HeyGen Video 1 had been added to Artificial Analysis's independent leaderboard and ranked #1 in text-to-video, with an Elo of 1160 on the AA-Video-T2V v2.1 board (which scores video-with-audio generation from pairwise human preference votes), and also topped the silent, no-audio version of that board at Elo 1144. That result was corroborated by independent AI-news coverage the same day.
It's worth knowing the sequence here. At launch on September 30, HeyGen Video had no independent score at all. As of October 1, it didn't appear on any of Artificial Analysis's three video boards, meaning every quality number HeyGen published in its own launch materials, based on an eval set of 4,800 votes, was self-reported. Third-party analysis flagged this gap explicitly at the time. The independent #1 ranking came about three weeks later, which does close that gap, but it's a useful reminder to check a model's actual leaderboard standing rather than taking launch-day marketing numbers at face value, for this release or any other.
Part of a bigger push, including a voice model
HeyGen Video didn't launch in isolation. Days earlier, HeyGen and real estate personality Ryan Serhant launched @HeySerhant on Instagram and TikTok, with a 12-month commitment to daily AI-generated content. And on October 22, HeyGen announced its voice model results alongside the video leaderboard news: HeyGen Voice Turbo now ranks #1 on the Artificial Analysis text-to-speech leaderboard, with the standard HeyGen Voice at #2.
For creators, that pairing matters more than it might seem. A top-ranked voice model and a top-ranked video model from the same vendor means visuals and audio can come from one API rather than three separate tools stitched together. HeyGen is not a small player here either: founded in 2020, it now serves over 40 million users across 196 countries, and 85% of the Fortune 100 use its platform, so this isn't a scrappy startup experiment, it's a scale product pushing into a new category.
What this means for your workflow
If you're producing short-form, multi-shot episodes, especially ones with a consistent cast and a need for varied cutaways, the practical appeal here is fewer tools. Instead of generating an image, animating it, writing dialogue, running lip-sync, and sourcing sound effects separately, HeyGen Video's pitch is that one prompt handles scene, lighting, and audio together. Combined with a top-ranked voice model from the same company, that's a single-vendor pipeline worth testing for b-roll and establishing shots.
That said, treat this like any new model launch: verify current leaderboard standing yourself before building a workflow around it, since rankings shift as more models get added and more votes come in. If you're assembling episodes with tools like Episodo, the useful move is to test HeyGen Video for the specific shot types where its one-call approach saves the most steps, short cutaways and stylized B-roll, rather than assuming it replaces your entire visual pipeline on day one.