From avatars to everything
HeyGen made its name on avatar video, where a script drives a presenter you already own. This is a different animal. The subject, the setting, the light, and the sound are invented from a prompt, not driven by a script. That puts HeyGen squarely in the general video-model race, Sora and Veo territory, though it arrives with an avatar company's customer list and an API-first distribution play rather than a consumer product.
The reference-to-video mode is the one to watch. For an avatar company whose clients are marketing teams and training departments, "keep this product, this person, this place" is the feature that sells the model. The price is what gets them in the door.
The price is the pitch
That pricing is the story for anyone who has watched per-second rates this year. Google's Veo 3.1 tiers run from $0.05 per second for Lite at 720p up to $0.60 for the full model at 4K. HeyGen's launch rate is roughly five to sixty times cheaper per second, depending on which Veo tier you compare against.
But the comparison deserves a discount of its own. 480p is a preview tier, not a delivery format. The resolution a social cut is usually finished at is 768p, and that costs 1.5 cents now, returning to 3 cents when the sale ends in October. The number that actually matters is the post-sale 768p rate, and it is the one HeyGen put furthest from the headline.
There is also a pattern worth naming. H3 is now a known quantity, and HeyGen is not claiming a new quality ceiling, nor does it need to. Its motion and prompt handling will resemble other H3 fine-tunes. The differentiation is cost and distribution. This is the second such play this week: Creatify's Boreal-H3 turned the same base model into an ad machine. Open video models have become a substrate. The business is whatever you build on top: the pricing, the workflow, and the customer list you already have.
What the price doesn't say
Three caveats travel with the headline. First, the launch rate has an expiry date: the end of October. After that, the real per-second cost of the usable resolution doubles. Second, API-only at launch means no casual tryouts. This is aimed at developers and businesses wiring it into a pipeline, not creators poking at a web interface. Third, audio generated in the same pass is convenient, but convenience is not quality. In-camera audio inherits whatever the base model can do, and H3's audio is competent rather than remarkable. For a cent a second, nobody is complaining. Just don't mistake the pricing for a technical claim it isn't making.
Sources
- [1] HeyGen developer documentation, "HeyGen Video" โ model modes, clip lengths, audio behaviour and MiniMax H3 lineageRead source
- [2] SLOP TV News, "HeyGen's first video model sells from 1 cent a second" (Oct 3, 2026)Read source