HeyGen Avatar Video

You need a person on camera and you don't have a person, a camera, or the time. This node hands you a HeyGen avatar - a pre-recorded digital presenter - and makes it say your script. Product demos, training videos, social clips, an explainer for a thing that doesn't exist yet: this is the node that produces the talking head without any of the production.

It's a HeyGen partner node, which means the rendering happens on HeyGen's servers through Comfy's API proxy and you pay per second of output with Comfy account credits. It landed in ComfyUI core in July 2026. Pick an avatar, type the script, get a video. That's the entire job.

How it works

HeyGen's avatars are real people (or generated characters) filmed against a background, driven by a text-to-speech engine or your own audio. The node packs your choice into a videos API call, polls until it's done, and hands back the rendered clip. The engine dropdown is where the real decisions live: it changes which avatars you can pick and how much each second costs. "auto" offers every avatar and picks its best engine - Avatar IV is the preferred default, Avatar V is the highest fidelity and priciest, Avatar III is the budget tier.

The speech input is a dynamic combo with two personalities. "script" gives you a text box plus voice settings (HeyGen TTS). "audio" lets you supply your own recording for the avatar to lip-sync - that's the one to reach for when the voice has to be a specific person, up to 10 minutes of audio.

The inputs that matter

• engine - "auto" or a specific rendering engine; each lists only avatars that support it.
• speech - script or audio, as above.
• custom_avatar_id - the escape hatch. Any of HeyGen's 3000+ public looks, or avatars you've created, by ID. Overrides whatever the dropdown picked.
• resolution, aspect_ratio, background_color - output size, and an optional solid hex background (leave background_color empty to keep the avatar's own footage).
• seed - purely a "force a re-run" button. It's not sent to HeyGen at all.

Output is a single VIDEO.

Gotchas

Watch the engine. Choosing Avatar V for "quality" doubles-ish your per-second cost, and not every avatar supports every engine - the node errors cleanly if you force a mismatch, so leave it on auto unless you know better. Backgrounds: a solid background_color looks clean and is handy for keying, but it replaces the avatar's set, so don't expect a green-screen miracle. And this is a cloud render with a per-second meter running - a 5-minute presenter video is a genuinely large spend, so render a short test take first and check the price badge before committing.