HeyGen Video brings scene generation and audio to its API
HeyGen Video generates complete scenes with sound from prompts and references. October launch rates depend on resolution and whether video references are used.
HeyGen announced HeyGen Video on September 30, expanding beyond avatars into generating complete scenes. The model is based on MiniMax H3 with additional training by HeyGen, according to the company’s original release announcement.
The launch catalog describes three ways to create a clip. Users can start with a written prompt, animate a first frame or supply reference material. Speech, background ambience and sound effects are generated with the video rather than added through a separate production step.
The lowest rate depends on resolution
HeyGen’s pricing explanation gives the October promotional rate for text and image generation as $0.01 per output second at 480p and $0.015 at 768p. The listed standard rates from November are $0.02 and $0.03 respectively.
Video references change the calculation. Promotional rates are $0.02 per billable second at 480p and $0.03 at 768p, with both input video and output video seconds counted. Image and audio references do not add reference charges. Comparing models using only the lowest advertised output rate would miss that difference.
The company says the model is available through its API and partners including OpenRouter, Runware and ComfyUI. The help page describes web access as forthcoming, so API availability should not be read as a promise that every HeyGen web account already has the feature.
Short clips and restrained scenes
The developer guide specifies clips lasting 5 to 15 seconds and output at either 480p or 768p. It supports up to nine image references, three video references and three audio references, subject to a combined maximum of twelve.
HeyGen’s own guidance favors a contained shot with one subject, one place and one action. It warns that extended text, delicate organic movement and detailed hand operations are less reliable. Long paragraphs and fine labels should be added as real typography after generation.
The benchmark is an internal evaluation
HeyGen presents a comparison based on 4,800 votes using prompts from Artificial Analysis Arena. The catalog labels that exercise an internal evaluation. It is not the same as an independently administered public leaderboard, and the headline quality claim should be read with that distinction intact.
For a production team, the useful test is whether approved clips preserve the product, obey the brief and need little repair. Measure the cost of usable output after retries and editing. Cheap seconds alone do not show how much a finished campaign will cost.
Featured image is an original AI generated conceptual editorial illustration.
