grok / v1.0 / image-to-video

Grok v1.0's image-to-video node is the one that turns a still you already like into motion. The workflow shape you probably want: generate your hero image in whatever local model you trust, wire it into this node's images socket, prompt the motion you want, and let Civitai's cloud run xAI's closed Grok model on it. You keep the image quality you can control and rent only the animation you can't run locally.

Like every node in civitai-comfy-nodes - Civitai's official pack of ~160 nodes generated from its Orchestration API spec - there's no local GPU involved. The node builds a videoGen job tagged grok / v1.0 / image-to-video, ships your image up as base64, long-polls until the job succeeds, and returns a native VIDEO tensor. Billing is in Buzz, and every second of output is metered.

The inputs that matter

prompt is the only required input. The stills ride in through images (an IMAGE socket - wire it from a Load Image, or feed a batch if you want the model to pick through them). Note it's plural: a list input, not a single image widget.

• duration - 1 to 15 seconds, default 6. Your cost lever, since video bills per second.
• resolution - 480p or 720p (v1.0 tops out at 720p; 1080p arrives with v1.5).
• aspect_ratio - auto by default, or pin 16:9, 4:3, 3:2, 1:1, 2:3, 3:4, 9:16.

Outputs: video (VIDEO), draft_cache_audio (AUDIO), draft_cache_video (VIDEO), workflow_id and raw_json (STRING). The node prints its workflow id and cost on itself after each run, which is how you catch a spend getting away from you.

Install and auth

Standard for this pack - official, Comfy Registry, one requirements file:

# ComfyUI Manager: search "Civitai Comfy Nodes"
cd ComfyUI/custom_nodes
git clone https://github.com/civitai/civitai-comfy-nodes.git
pip install -r civitai-comfy-nodes/requirements.txt

Restart and authenticate via a Civitai Auth node, the CIVITAI_API_TOKEN env var, or the sidebar OAuth login.

Where people get burned

Cost first. Image-to-video jobs are billed per output second, and if your image doesn't move the way you wanted, the retry is billed too. Iterate on the prompt at the shortest duration before committing to long clips. Your image leaves your machine - it's uploaded to a third-party server, full stop; don't feed it anything you wouldn't send to xAI directly. Early preview. The pack README warns nodes and behavior "may change without notice." And one version quirk worth knowing: v1.0 uses the kebab-case operation naming (image-to-video) while the v1.5 nodes switched to imageToVideo - same engine family, but don't be confused when the node names don't quite match each other. If you want multiple reference images driving style rather than one start image, the v1.5 referenceToVideo node is the upgrade path.