HeyGen launches Video 1 model with API at one cent per second

The model in HeyGen's post is what HeyGen elsewhere calls HeyGen Video, its first general-purpose video model. Until now the company was known for avatar and voice models; this one generates a whole scene with no avatar, from a text prompt, a starting image, or a reference video, and produces dialogue, ambience and sound effects in the same call. It is not brand new: HeyGen's help center article was last updated September 30, and OpenRouter's listing, which HeyGen names as its primary launch partner, gives the same release date. The help center calls the model built in-house, while HeyGen's September 30 launch post said it was built on MiniMax's H3 and post-trained by HeyGen.

The headline price needs its conditions. According to the help center's pricing table, HeyGen is running a 50% launch promotion through the end of October, and $0.01 per second is the promotional rate for text-to-video or image-to-video at 480p. The same modes cost $0.015 per second at 768p, and the listed standard rates from November are double: $0.02 and $0.03. Reference-to-video costs more and bills both the input and output seconds. The post pairs the one-cent figure with a 2K shot, but the help center table lists only 480p and 768p; OpenRouter's listing, as of October 22, adds a 2K tier and quotes a range from $0.01 per second at 480p to $0.05 per second at 2K. In other words, one cent buys a second of the cheapest, lowest-resolution output while the promotion lasts.

The speed and quality claims are HeyGen's own. In the narration of the launch video, as captured by an automatic transcript that does not cover the visuals, HeyGen says most of the shots shown came back in under a minute, a softer statement than the post's. The help center says HeyGen Video ranked first for quality in blind side-by-side comparisons against Seedance 2.5, Seedance 2.0, H3 Max and Kling 3.0, from 4,800 blind votes on the Artificial Analysis Arena query set. That describes a comparison HeyGen reports using Artificial Analysis's prompts, not a published Artificial Analysis leaderboard placement. OpenRouter lists clips of 5 to 15 seconds.