Grok Imagine Video

Grok was a text model, then an image model, and now xAI ships video - and thanks to this Runware pack node, you can run it from ComfyUI without owning any of the hardware it would take. The Grok Imagine Video node is a thin client for xAI's video model: you type a prompt (or hand it images, frames, or existing footage), it runs in the cloud, and a native VIDEO comes back.

The genuinely useful part of this model is that it's not just text-to-video. The referenceImages socket lets you anchor a subject, frameImages lets you define keyframes the clip animates between, and the video input takes a URL/path/UUID of footage you want to edit. So you can take a generated image from elsewhere in your graph, feed it here, and ask for motion on that exact subject instead of rolling the dice on a fresh generation.

What you set

Only positivePrompt is required, and the size dropdown is the first thing you'll actually touch - it's a preset list (480p and 720p in a bunch of aspect ratios, plus "from input" options that respect your source) rather than raw width/height. That's a deliberate design choice: video models are picky about resolution buckets, and the presets are the ones that work.

• size - pick your bucket. 480p is cheap and fast, 720p if the shot needs detail.
• duration - a toggle (flip it on) plus duration_value in seconds, 1–15. Longer isn't always better; 15 seconds is a lot of temporal coherence for any model to hold, so shorter clips tend to look tighter.
• frameImages / referenceImages / video - the media inputs described above.
• numberResults - up to 4 variations per queue.

The safety.mode dropdown (none/fast/full) is worth a look if the model's built-in moderation is fighting you or, conversely, if you're generating client content and want the belt-and-braces check. Output is a single video.

The one setting beginners trip on

This pack gates its model parameters behind on/off toggles - duration is a boolean, and the real value lives in duration_value, which stays greyed out until you flip the switch. Same pattern as the other video nodes here. Want an 8-second clip? Toggle duration on, then set duration_value to 8. Typing the number first does nothing.

Install

It's the same story as every node in this pack:

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

Or install "Runware" from ComfyUI Manager and restart. You'll need an API key from the Runware dashboard (Settings → Runware API key, or the RUNWARE_API_KEY env var), and video is the priciest thing in this pack - the node shows the per-run cost on its title bar, so keep an eye on it. There are no model files to download; the entire model lives on xAI's side, which is the appeal and also why there's no offline mode.