zhenzhen-grok-video-tools-fal

Most video nodes start from nothing - text in, clip out. This one starts from a clip you already have, and does one of two jobs: extend_video (keep the moment going past where it stopped) or edit_video (re-shoot it into something else, like "colorize" or "make it claymation"). It's the xAI Grok Video Tools endpoint wrapped up as a ComfyUI node, and it's the natural next step after a Grok text-to-video node gives you a clip you like but want longer or different.

Like every node in the Comfyui-zhenzhen pack (T8mars' fan-priced API reseller, ai.t8star.org), this is a pure API wrapper. No local model, no GPU work on your side, per-call credits, and your api_key lives in the node widget. The -fal in the name means the request goes through Zhenzhen's proxy of fal.ai's queue API - submit, poll, download - and billing is pre-deducted with the difference refunded after the job lands.

How it works

You give it a video, it gives you a video. The video comes in through either the video input (a VIDEO tensor from elsewhere in your graph - say, a previous Grok generation or a locally-loaded clip) or a video_url string, with video_way choosing upload vs video_url. The mode switch picks the endpoint:

• extend_video - continuation. duration matters here (1–15s, default 6): how many seconds to keep going.
• edit_video - restyle. resolution matters here (auto/480p/720p), and the prompt is the instruction - the default is literally "Colorize the video."

The result is returned as a VIDEO output you can save directly, plus video_url and response strings for grabbing the file or debugging.

Inputs that matter

• prompt - the instruction. For extend it's usually short or empty; for edit it's the whole job.
• mode - extend or edit; pick the one you actually want, because duration and resolution swap meaning between them.
• video / video_url - wire one. If video is connected, video_url is ignored.
• poll_interval / max_poll_attempts - patience controls; the 600 × 6s default is a one-hour ceiling, generous for these jobs.

Installing it

Install the pack once via ComfyUI Manager (search Comfyui-zhenzhen) or:

cd ComfyUI/custom_nodes
git clone https://github.com/T8mars/Comfyui-zhenzhen

Restart ComfyUI, paste your key into the widget. There are no model downloads - the "dependency" list in the repo is mostly the standard ComfyUI stack plus openai/opencv-python for the pack's other nodes.

Common issues

• "API key not provided" - the widget is empty; fill it. There's no global key file anymore, keys belong in the node.
• Connection / 443 errors - overseas servers; Zhenzhen's README expects you to run a VPN in TUN mode and to disable junk antivirus/router filtering. The node auto-fails over between .org and .cn.
• 500 errors - usually the upstream vendor choking, not you. The pack's own docs: just run it again.
• Extend producing something weird - a one-line or empty prompt gives Grok very little to work with. For edit_video especially, put the style word in the prompt ("restyle as watercolor") and let the default resolution: auto keep the source's look.

If your pipeline is "generate with Grok → tweak with this node," you're using the pack the way it's designed: each step is a separate, individually-priced API call, and you only pay for the passes you actually need.