HeyGen Avatar IV

This is the "AI presenter" node. Give HeyGen Avatar IV an image of a person and some speech, and it comes back with a video of that face talking, lipsynced to the audio - the thing you've seen a thousand times in corporate onboarding videos and, lately, in plenty of very convincing deepfakes. It's the rare ComfyUI node that isn't about generating media at all; it's about animating someone. That makes it useful and also worth a beat of caution about what you're animating and who owns the face.

It runs through the Runware/ComfyUI-Runware pack, which means no local GPU and no HeyGen account to set up separately - one Runware API key covers this node and every other model in the catalog. The heavy lifting happens on HeyGen's servers; your graph just passes parameters.

How it works

The node takes either a script to be spoken (via the Runware Speech builder, which carries text plus a voice pick) or an audio file (UUID or URL) to lipsync against, then drives HeyGen's avatar API with an avatar ID. That dropdown is the star: 514 preset avatars, all the named photoreal stock presenters plus your own photo/video avatars if you've trained them on HeyGen's side. Output comes back as a native VIDEO (or a saved file path when ComfyUI's runtime lacks the native type), and the title bar shows the run cost like every other node here.

The inputs that matter

• speech - wire in a Runware Speech builder node with the text and a voice. This is how you get dialogue without preparing an audio file yourself.
• audio - the alternative path: an existing audio clip as a UUID or URL. Pick one, not both, depending on whether you're scripting or reusing a track.
• avatar - the presenter. Browse the 500+ presets or use (default). If you want a specific person from an image, that's what the image input is for (HeyGen photo avatars).
• image - your own still to animate, if you're not using a stock avatar.
• background (IMAGE) or settings.backgroundColor (hex) - where they stand. The background socket takes an actual image; the hex field is for a solid color. settings.removeBackground exists too, with the honest caveat baked into the tooltip: video avatars must have been trained with matting enabled for it to work.
• settings.expressiveness - low/medium/high. Low keeps things stiff and professional; high gets animated eyebrows.

size (720p/1080p), numberResults (max 4), outputFormat, and the standard safety/TTL/output-quality fields round it out.

Installing

Install the pack once and every node in it shows up together:

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

Restart, or use ComfyUI Manager and search Runware. Needs a Runware API key - Settings panel, RUNWARE_API_KEY, or runware auth login.

Where people get burned

The speech socket is a custom type, not a text box - beginners drag it in expecting a prompt field and get a dangling input. Add the Runware Speech builder from Runware/Params. Also, don't expect a deepfake-level tool from the stock avatars; the preset presenters look like stock presenters, which is the point. If you want your own face moving, that means training a photo avatar on HeyGen's platform first - the node is the runtime, not the trainer. And be smart about consent; this is a node that makes it very easy to put words in someone's mouth, and that cuts both ways.