Seedance 2.0 for multi‑character AI video

Most artificial intelligence video tools can fake one person talking to a camera, but they usually break when two people need to share a scene, keep the same faces, and move like they have actual weight. HeyGen said on April 7, 2026 that Seedance 2.0 is now built into its platform, and its product page says multiple verified digital twins can now appear in one cinematic scene. (heygen.com 1) (heygen.com 2) That is the jump from “talking head” video to something closer to a directed commercial. HeyGen says the tool can place a verified digital twin into wide shots, close-ups, crane shots, and moving camera scenes while keeping the same face, clothes, and identity across cuts. (heygen.com) A digital twin is the company’s term for an avatar trained on real footage of a real person, not a face invented from a prompt. HeyGen’s developer documentation says enterprise users create one by uploading at least two minutes of 720p video plus a separate consent video that explicitly grants permission to build the avatar. (docs.heygen.com) HeyGen has been pushing that identity angle hard because Seedance itself does not allow real human faces through its public application programming interface. HeyGen says its own consent and identity-verification system is the reason a customer’s verified face can appear inside Seedance-generated footage when “no other platform can offer this.” (heygen.com 1) (heygen.com 2) The other half of the story is InVideo, which is selling Seedance less as an avatar product and more as a video engine inside a menu of models. Its Seedance page says users can generate multi-shot videos from text and images, edit the result with text prompts, and export in 1080p high definition. (invideo.io) InVideo’s pitch is that Seedance is no longer trapped in the old “12-second clip” box that defined a lot of early artificial intelligence video demos. The company says its “Seedance Pro” workflow can create videos of any length, with connected scenes, stable motion, and camera moves like pans, zooms, and tracking shots. (invideo.io) HeyGen is packaging the same underlying model in a different way. Its integration page says Seedance 2.0 now powers three tools inside HeyGen, including Avatar Shots for cinematic scenes with a verified digital twin and Video Agent for prompt-to-video generation up to three minutes long. (heygen.com) That matters because the hard part in synthetic video is no longer just making one pretty shot. The hard part is keeping one character coherent from the first frame to the last, and HeyGen’s newer Avatar V system says it can build that consistency from a 15-second webcam recording and preserve it across angles, backgrounds, and languages. (heygen.com) Put those pieces together and the new product looks less like a toy generator and more like a small video studio in software. One company is combining a face you have verified, a script you type, and camera motion you describe, then promising a finished ad, lesson, or social clip without a crew, a set, or a reshoot. (heygen.com 1) (heygen.com 2) (invideo.io) The immediate buyers are likely marketing teams, agencies, and creators who already make lots of short-form video and care more about speed than film-school perfection. InVideo says Seedance 2.0 is available on its Max, Generative, and Team plans, while HeyGen is presenting Seedance as part of its broader video stack rather than a separate research demo. (invideo.io) (heygen.com) What changed this week is not that artificial intelligence can generate video at all. What changed is that two consumer-facing companies are now shipping multi-character, identity-locked, camera-directed video generation as a normal product feature, which is usually the moment a research trick starts turning into everyday creative software. (heygen.com) (heygen.com) (invideo.io)