Grok Imagine Video 1.5 Preview

What it is

The newer Grok Imagine video model, at KIE Next / Video / xAI / Grok Imagine, and the one I'd try first in this folder. It does text-to-video and image-to-video in a single node - the image socket is optional - takes a sensible second-level duration (1 to 15 seconds) instead of the older nodes' blunt 6-or-30 menu, and gives you auto aspect ratio so the model can infer the frame from your prompt or your source image.

"A preview" in the name is honest: this is xAI's working version, and the pack generates it as its own node rather than hiding it behind a dropdown, which means a workflow pinned to it keeps working when the next version lands as a different node.

How it works

Prompt and optional images go up to KIE, the task is submitted, the node polls it to completion, then the finished MP4 is downloaded and handed back as a native VIDEO object. Nothing runs locally; the compute is on KIE's side and the credit is spent at submit time.

If you connect an image, the pack uploads it (PNG) and passes the URL. Without one, the node is a pure text-to-video generator. Both modes are the same endpoint, which is why you don't need two nodes here.

The inputs

prompt is required - up to 4096 characters, and unlike the older text-to-video node, there's no hard motion-craft guidance in the docs; describe the shot however you like.

• image_urls - the optional IMAGE socket. KIE documents this as accepting multiple uploads (up to seven files, 20MB each), with only one supported when you ask for 1080p. Feed a single image for image-to-video; feed several and you're in reference territory.
• duration - an INT, default 8, range 1–15. This is the flexibility the older Grok video nodes don't have, and it's the main reason to prefer this one.
• resolution - 480p (default), 720p, 1080p. The default is cheap on purpose; go up once the shot is right.
• aspect_ratio - auto (default), 1:1, 16:9, 9:16, 3:2, 2:3. The tooltip carries a real caveat: with a single image the parameter is ignored and the clip inherits the source frame's shape.
• nsfw_checker - off by default, with the same plain-spoken caveat as the rest of the family.
• timeout_seconds (1200) and callback_url (leave it empty for normal ComfyUI use).

Outputs

video - a ComfyUI VIDEO you can save, preview, or post-process. url is the primary result URL; all_urls_json is every URL the task returned; task_id is the handle you want if you plan to extend or upscale the clip with the pack's other Grok Imagine video nodes; raw_json is the provider response for when something fails and you need the real reason; credits_consumed and credits_left report cost and balance and are also printed onto the node after each successful run.

Install

Manager → KIE.ai Nodes Next. Manually:

cd ComfyUI/custom_nodes
git clone https://github.com/felipederosilva/ComfyUI-KIE-Nodes-Next
python -m pip install -r ComfyUI-KIE-Nodes-Next/requirements.txt

The requirements are requests, Pillow, numpy, PyYAML; there are no weights to download because the model is hosted, which is the whole point of this pack. Restart ComfyUI, then complete the one-time key setup at Settings → KIE.ai Nodes Next → Connection (a KIE_API_KEY environment variable also works and takes precedence). One install quirk: the pack generates its node list from KIE's live docs index on first launch and re-syncs at most daily, reloading the ComfyUI frontend once during that first sync. If the node is missing, refresh the catalog rather than reinstalling.

Gotchas

• Video jobs are where API spending gets real. A 15-second clip at 1080p is not the same purchase as a 480p 8-second test. Draft low, finish high.
• Called "preview" for a reason. Behaviour can shift when xAI ships the next version; yours stays on the node you pinned, but the model behind it may not be frozen forever.
• Aspect ratio is ignored for a single image. Don't fight it - crop or letterbox downstream.
• VIDEO output needs a current ComfyUI build. The pack uses comfy_api's file-backed video helper and tells you to update if it's missing.
• No key gives No KIE API key configured... - the Settings step, not your wiring.
• Compare before you commit. For repeatable shots, a local Wan or LTX pipeline is free per run. This node's edge is a good-looking 12 seconds without a GPU budget.