HeyGen's first general-purpose video model — it generates a whole scene (picture, dialogue, and sound effects) from a text prompt, plus image-to-video and reference-to-video, at up to 2K.
Last verified · 2026-10-25 · by Moe Ameen
HeyGen Video is HeyGen's first general-purpose video model, introduced in late September 2026. HeyGen is best known for avatar video — a scripted talking-head presenter — and HeyGen Video is a deliberate departure: it generates a whole scene, not an avatar. From a single text prompt it produces the picture, the dialogue, and the sound effects in one pass, so the clip arrives with audio instead of as a silent video you score afterward.
The API (model ID `heygen-video-1`) exposes three modes: text-to-video, image-to-video, and reference-to-video, the last letting you supply your own images, clips, or audio to steer the look. Output runs 480p up to 2K as MP4 (H.264, 24fps with AAC audio), in aspect ratios including 9:16, 16:9, 21:9, 1:1, 4:3, and 3:4 (1080p and 2K are limited to 16:9 and 9:16), with clips from about 5 seconds up to roughly 15–30 seconds depending on the mode and quality setting. Styles span live action, anime, claymation, and screen print, and prompts can direct camera, lighting, and sound. HeyGen's own developer docs say the model is built on MiniMax's open-weight H3 base and post-trained by HeyGen.
What put it on the map is the leaderboard. HeyGen says HeyGen Video ranked first in a blind text-to-video test on Artificial Analysis; more precisely, it leads Artificial Analysis' silent board — which judges the picture with audio turned off — with an Elo around 1144, on blind human-preference votes, and sits a few spots back (not first) on the audio-on board, behind Wan 3.0 and Utopai X. HeyGen priced it aggressively at launch: a 50%-off promotion through October puts 480p at about $0.01 per second and 768p at about $0.015 per second, both with audio; standard rates return to roughly $0.02 and $0.03 per second respectively in November, with 1080p/2K billed at about three times the 768p rate. The HeyGen app is free to start with a couple of clips a day.
HeyGen names the model's weak spots plainly — long on-screen text, close-up hand work, and soft organic motion like petals, paper, and hair — and says it is strongest on short, contained shots with one subject, one place, and one action. Treat the exact prices, ranking, and specs as a snapshot and confirm them against HeyGen's own pages before building on them.
HeyGen Video's most underrated feature for a brand is reference-to-video: feed it your own images, clips, or audio and it holds that look across generations, so a run of scenes reads as one visual identity instead of a new world every render. That consistency is exactly the raw material a recurring series needs — but HeyGen Video still only ever hands back a short clip with no caption track, no brand frame, and whatever single aspect ratio you asked for. [Kompozy](/) is where a consistent look becomes a recurring, on-brand, published format.
Generate a set of reference-consistent scenes through HeyGen Video, then drop them into Kompozy. [Clipped Shorts](/glossary/clipped-short) takes an uploaded clip and cuts it to vertical with word-synced captions — the track every clip needs because feeds autoplay muted, even with HeyGen Video's generated audio. Because the visual identity is already locked by your reference, Kompozy layers the brand on top: a [Carousel](/glossary/hyperframes) built with pixel-exact HyperFrames, quote graphics of the key line, per-platform text posts, a blog, and a newsletter — all held to your [Persona Brief](/glossary/persona-brief) so the copy stays on voice while HeyGen Video keeps the scene on-model. Kompozy's own face-locked persona pool then carries a talking-head presenter into [Persona Shorts](/glossary/persona-shorts) and [Persona HeyGen](/ai-tools/heygen) between scene renders, so a HeyGen Video establishing shot can cut straight into an avatar segment. [Autopilot](/glossary/autopilot) and a per-post review pipeline schedule and publish the whole set across eight social platforms plus blog and email. HeyGen Video holds the scene for a few seconds; Kompozy turns it into a weekly branded show.
HeyGen Video (API model ID heygen-video-1) is HeyGen's first general-purpose video model, introduced in late September 2026. It generates a whole scene — picture, dialogue, and sound effects — from a single text prompt, and also supports image-to-video and reference-to-video, at 480p up to 2K.
No. HeyGen is known for avatar video (Avatar V, Avatar IV, the Video Agent), but HeyGen Video makes no talking-head presenter — it is a general-purpose scene generator. HeyGen's own developer docs say it is built on MiniMax's open-weight H3 base and post-trained by HeyGen, and it tops Artificial Analysis' silent text-to-video leaderboard in blind testing.
HeyGen set a 50%-off launch price through October: about $0.01 per second at 480p and about $0.015 per second at 768p, both with audio. Standard rates return to roughly $0.02 and $0.03 per second respectively in November, with 1080p/2K billed at about three times the 768p rate. The HeyGen app is free to start with a couple of clips a day. Confirm live rates on HeyGen's pages, since the headline price is promotional.
HeyGen flags three weak spots: long on-screen text, close-up hand work, and soft organic motion like petals, paper, and hair. Clips are also short (about 5–30 seconds), and 1080p/2K output is limited to 16:9 and 9:16. It is strongest on short, contained shots with one subject, one place, and one action.
Kompozy is the finishing and distribution layer. Generate scene clips with HeyGen Video, upload one to Kompozy's Clipped Shorts for captioned vertical cuts, and spin the idea into carousels, quote graphics, per-platform text, a blog, and a newsletter in your Persona Brief voice — then publish across nine destinations: eight social platforms plus blog and email.