grok / v1.0 / edit-video

Of all the video nodes in civitai-comfy-nodes, this is the one that breaks the pattern: instead of a still or a prompt, its required input is a video. Grok v1.0's edit-video mode takes an existing clip, understands what's in it, and re-renders it to match a new instruction - restyle the scene, change the action, swap the mood - and returns a fresh clip. If you have footage you want re-cut without hand-editing, this is the node.

The pack is Civitai's official ~160-node set, generated from the Orchestration API spec, and this node is a cloud call like the rest: it submits a videoGen job tagged grok / v1.0 / edit-video, uploads your source clip, long-polls until the job succeeds, and returns a native VIDEO tensor. Billed in Buzz. Nothing local.

What you set

• prompt (required, multiline) - the edit instruction: "make it night", "change the camera to a slow orbit", "the character now wears a red coat."
• video_url (required) - a VIDEO input. Wire it from a video loader/import or anywhere else in your graph that produces the pack's native VIDEO type. Despite the name it's a real socket, not a URL text field.
• duration - 1 to 15 seconds, default 6. Output length; billed per second.
• resolution - 480p or 720p (v1.0 ceiling again).
• analyzed_duration - how many seconds of the source clip the model should actually look at. Default 0, which usually means "let the service figure it out." Set it when you only want the first N seconds of a long clip analyzed.

Outputs: video (VIDEO), draft_cache_audio (AUDIO), draft_cache_video (VIDEO), workflow_id and raw_json (STRING).

Install and auth

Standard pack install - official, Comfy Registry:

# 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, authenticate via a Civitai Auth node, CIVITAI_API_TOKEN env var, or sidebar OAuth.

Common issues

Upload time and cost both scale with the source clip. An edit-video job ships your footage to Civitai's cloud, and that upload plus the re-render is what you're paying for - trim your source to the part you actually want edited before wiring it in. analyzed_duration is the knob people skip. If your clip is long and the edit drifts or ignores the later half, set it explicitly rather than hoping the default guesses right. Your footage leaves your machine - it's uploaded to a third-party server, so respect that in what you feed it. And the usual pack caveats: Buzz spend accumulates fast on video, and the pack is early preview - nodes and behavior "may change without notice." Also worth knowing: v1.0 is the only Grok version here with a true video-in video-out edit node; the v1.5 menu swaps toward reference-image conditioning instead.