# LivePortrait / ComfyUI

A talking portrait you run. LivePortrait is MIT research code; ComfyUI is GPL-3.0 — Synthesia and HeyGen without the hosted avatar seat.

**Replaces:** Synthesia / HeyGen

**Tier:** Medium

## Problem

Synthesia and HeyGen sell hosted avatar video from a script. You wanted a portrait animator on hardware you control. LivePortrait is MIT (KlingAIResearch/LivePortrait): a research model that drives a still portrait from a video. ComfyUI is the node graph that can host that kind of workflow. Piper can speak the script (/replace/elevenlabs). This is not Synthesia's stock avatar library, HeyGen's translator SKU, or a turnkey studio. You still need a GPU and you still bring faces and audio you have rights to.

## Stack

- [LivePortrait](https://liveportrait.github.io) — MIT research portrait animator, not a Synthesia studio
- [ComfyUI](https://www.comfy.org/) — GPL-3.0 node graph that can run video/portrait workflows
- [Piper](https://github.com/OHF-Voice/piper1-gpl) — GPL-3.0 local TTS for the script

## Estimated SaaS cost saved

| You stop paying | Typical / year | Self-host instead |
|---|---:|---|
| Synthesia / HeyGen (small team) | $1,200 | LivePortrait + ComfyUI + Piper |

**About $1,200 / year** on a small team, before you count lock-in.

## Deploy on Railway

A GPU workstation is the honest path. LivePortrait and ComfyUI want weights on disk. No one-click avatar studio in this repo — Start on Railway. Still images stay on /stacks/comfyui-image. Speech stays on /stacks/piper-tts.

[Start on Railway](https://c17.io/go/railway/liveportrait-avatar?from=%2Fstacks%2Fliveportrait-avatar.md)

[ComfyUI if this is still images, not a talking portrait](https://c17.io/stacks/comfyui-image)

## AGENT_PROMPT

```
Replace Synthesia or HeyGen with LivePortrait (portrait animation) plus ComfyUI if you want a node graph, and Piper for speech.

- Read LivePortrait's model card. Animate one still you have rights to. Do not claim Synthesia avatars or HeyGen translation.
- Image generation stays on /stacks/comfyui-image. Write AGENT_PROMPT.md.
```

## FAQ

### Is this a Synthesia avatar library?

No. You bring a face and a driving video or audio. The stock avatar catalog is the product this page leaves.

### Synthesia or HeyGen?

Same published stack. LivePortrait is the animator. ComfyUI is the graph. Piper speaks.

### Why not just ComfyUI?

ComfyUI is still images and whatever video nodes you add — /replace/midjourney, /replace/runway. This page is the talking-portrait job.

Source: https://c17.io/stacks/liveportrait-avatar