Concurrent Submit | Zhenzhen Grok3 Video

If you're new to Grok video through the Zhenzhen pack, this is the node to start with. It's the concurrent wrapper around the plain Grok3 Video API node, and it's the balanced option: 6/10/15-second clips, real resolution choices (480P/720P/1080P), audio built in, and multi-image reference support. The "30S" sibling gives you length at 720P only; this one gives you the resolution ladder.

And because it's a Concurrent Submit node, you can fire ten of these at once into the pack's shared video pool instead of watching ComfyUI block on one clip at a time. That matters a lot for video, because a single Grok clip can take a while to render - batching is how you make the workflow feel fast even when the backend is slow.

What it does under the hood

Underneath, the plain ComflyGrok3VideoApi node uploads your reference images to a files endpoint, posts the prompt to Grok's video API through Zhenzhen's proxy (ai.t8star.org), then polls until the clip is ready. The Submit wrapper doesn't change any of that. It just defers it: the call runs on a background worker in the shared video pool (default 10 workers, COMFLY_VIDEO_CONCURRENCY env var), and the node returns a task (COMFLY_VIDEO_FUTURE) instantly instead of the video itself.

So the output - a single task - is only useful when it lands in ComflyConcurrent_Video_Await. That collector blocks until all ten slots are done, restores your submission order, and hands back video_1 through video_10 plus a status JSON. failure_mode on the collector decides whether one bad clip aborts everything (fail_fast) or leaves a placeholder slot.

[Submit: Grok 1080p shot A] ──task──┐
[Submit: Grok 1080p shot B] ──task──┼──▶ ComflyConcurrent_Video_Await ──▶ Save
[Submit: Grok 1080p shot C] ──task──┘

Inputs that matter

prompt is required; Grok handles Chinese prompts well and generates matching audio, which the pack markets as a Sora2 substitute for dialogue-heavy short drama. duration is 6, 10, or 15 seconds (default 15 - the pack's changelog notes 15s is the fan-favorite length for character pieces). resolution defaults to 1080P; drop to 480P when you're just testing ideas and want cheap fast renders. ratio defaults to 1:1 with 2:3, 3:2, 16:9, 9:16.

image1–image7 enable image-to-video and multi-reference, with the @img1 … @img7 syntax in the prompt to say which reference does what. seed is cache-control only (Grok doesn't expose a native seed), and api_key is optional once the pack's API Settings node is configured.

Install and the usual suspects

Install is the standard pack story: ComfyUI Manager → search Comfyui-zhenzhen, or git clone https://github.com/T8mars/Comfyui-zhenzhen into ComfyUI/custom_nodes and restart. No models to download.

The failure modes are worth knowing before you hit them. A 500 is usually Grok/upstream overload - the pack's update log treats these as "run it again," and failed tasks generally don't charge. A 429 is the same story at the rate-limit level. If a finished clip reports failed at the Await node, check the site's async-task page: your video is probably there, and the poll just expired. And remember the resolution ladder costs real money per render - 1080P and 15s add up fast, so use the concurrent batching for drafts at 480P/6s and save the expensive settings for keepers.