HeyGen Avatar V

HeyGen is the avatar-talking-head company - the one that makes a photoreal presenter say any script in any language with matching lip sync. Avatar V is their current generation of that, and this node drops the whole thing into ComfyUI: pick an avatar from a dropdown, give it something to say, and get a VIDEO of that presenter talking. It's the one node in this pack that turns your graph into a content-marketing studio.

If you've ever needed "a professional-looking person to read my script on camera" and lacked the budget, the talent, or the lighting rig, this is the node. It's the corporate-video and training-content play, and it's why HeyGen's commercial business exists. It runs entirely on Runware's cloud - there are no local weights, no "trained avatar" files on your disk, and no way to run it offline.

Inputs that matter

The avatar dropdown is required, and it's a genuinely wild list: 500 photo and video avatars named like Abigail_sitting_sofa_front, Amelia_standing_business_training_side, and so on - poses and settings baked into the choice. Pick a name and you've picked the presenter, the camera setup, and the vibe.

Then the inputs that turn a static avatar into a talking one:

• speech - a RUNWARE_SPEECH value from the Runware Speech builder (under Runware/Params). That builder takes text (the script; it supports speaker tags like [Sam] Hello [Bob] Hi there and markup like [laughs]) and a voice identifier, plus volume/speed/pitch/language. This is how you say what the avatar says and how.
• audio - the alternative: a UUID/URL of a pre-made audio track. Use this when the voice is already recorded and you just need the avatar to match it.
• background - an IMAGE to place the avatar in front of, or use settings.backgroundColor (hex, e.g. #ff0000) for a solid backdrop.
• settings.removeBackground - cut the avatar's background out (requires video avatars trained with matting). Pair with a background image for the "insert presenter into my footage" workflow.
• settings.fit - cover (fill the canvas, may crop) vs contain (fit entirely, may show background). Default lets the server pick.
• settings.caption - burn captions into the video and return a sidecar SRT. The accessibility no-brainer for social clips.
• size - resolution/aspect dropdown up to 4K ("from input" variants included).

numberResults (up to 4), outputFormat (MP4/WEBM/MOV), outputQuality, and safety round it out.

How it works

The node sends heygen:avatar@5 as a videoInference task through the Runware SDK; the speech, voice, and avatar ID travel in the request. The finished presenter clip comes back as a native VIDEO socket for SaveVideo. Cost and content-check status print on the title bar - and with a 500-item avatar catalog, "which avatar" is the expensive choice, so preview before you commit to a full-length render.

Install and gotchas

cd ComfyUI/custom_nodes
git clone https://github.com/Runware/ComfyUI-Runware
pip install -r ComfyUI-Runware/requirements.txt

Restart; API key via Settings → "Runware API key", RUNWARE_API_KEY, or runware auth login. Paid per render, minimum top-up.

The traps: forgetting that the script lives in the Runware Speech builder, not on this node - you must wire speech in, or nothing gets said. And the avatar list is fixed; you can't upload your own face with this node (that's HeyGen's custom-avatar product, a different integration). Also, avatar talking-head video is not cheap, so keep caption and audio decisions settled before you queue long scripts. If "AI presenter reads my script" is a workflow you run often, this is the node - just treat the avatar catalog like a casting call, because that's literally what it is.