P-Video-Avatar

P-Video-Avatar is a talking-head generator from Pruna AI, and it does one very specific job: take a single image of a person and animate them speaking. You feed it a face, tell it what to say (and how to say it), and get back a video where the mouth moves in sync with audio. If you've been assembling the same result out of a local I2V model plus a separately-synced voice track, this collapses the whole pipeline into one node.

It runs on Runware's cloud, so the VRAM question never comes up - Pruna's models are typically optimized for fast, cheap inference, which is the whole brand. The trade is the usual one for this pack: it's a billed API call per generation, and everything is sent to their servers.

What you set

frameImages is required - one or more IMAGE inputs of the person to animate. After that:

• speech - a RUNWARE_SPEECH input with the text to say (this is a builder-socket type; a text string alone isn't enough, you'll want the Runware speech/params node upstream).
• audio - alternatively, a URL or UUID of a real audio track to lip-sync against, if you already have the voice recorded.
• settings.voicePrompt - delivery instructions: tone, pacing, emotion. Defaults to "Say the following." so your positivePrompt text becomes the script. This is how you get a flat read vs. an excited one.
• resolution - 720p or 1080p; when you feed input media it auto-matches the input's aspect ratio.
• seed - exposed here, so unlike some Runware video nodes you can reproduce a take. Set it, keep numberResults at 1, and the output is deterministic.

settings.promptUpsampling rewrites your prompt for better results, and settings.safetyFilter (default on) screens the prompt and input image before generation - turn it off only if you're hitting false flags on legit content.

Output is video (VIDEO socket) - wire it into a video preview/save node.

Installing

This is one node in the Runware/ComfyUI-Runware pack; install the pack and it appears under Runware/Video/prunaai.

ComfyUI Manager: search Runware, install, restart. Manual:

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

No models to download (runware-sdk, pillow, soundfile are the whole dependency list). You need a Runware API key - runware.ai/api-keys, then ComfyUI Settings → Runware API key or RUNWARE_API_KEY.

Where people get burned

Two things trip people up. First, the required speech socket is easy to ignore - pasting text into positivePrompt alone doesn't drive the avatar unless it's wired through a RUNWARE_SPEECH input or you supply audio. Second, don't expect a documentary-grade clone: avatar models like this are great for explainers, characters, and quick dubs, and noticeably less great at preserving a specific real person's exact mannerisms. Also, since the whole thing runs remotely, don't feed it anything you wouldn't upload to a cloud service.