Concurrent Submit | zhenzhen-grok-video-1.5-fal

This is the "test cheap before you spend" Grok video node. It wraps zhenzhen-grok-video-1.5-fal, the FAL-routed Grok 1.5 video endpoint, and the single most useful thing about it is the duration tooltip: 5 seconds is the lowest-cost test duration. For a dollar-conscious workflow, that's the whole pitch - validate your prompt and framing with a 5s clip at 480p, then go big.

It's a concurrent submit node too, so you can run ten of those cheap tests at once in the pack's shared video pool instead of waiting for them one by one. The FAL route here means billing is pre-paid/deposit-based (the pack's README describes FAL models as 预扣制 - a hold is taken up front and settled after the render), so a fast batch of 5s tests is also the cheapest way to learn what a given prompt actually costs.

How the pieces fit

The plain Comfly_grok_video_1_5_fal node submits to a FAL queue endpoint through Zhenzhen's proxy and polls until the clip is ready. The Submit wrapper keeps every input, swaps the output for a task (COMFLY_VIDEO_FUTURE), and runs the call on a background worker in the video pool (default 10 workers, COMFLY_VIDEO_CONCURRENCY env var).

The task output is only useful when connected to ComflyConcurrent_Video_Await. That node waits for all ten slots, restores order, and returns video_1 … video_10 plus a status JSON. If you'd rather one failed clip not kill the whole batch, set failure_mode to placeholder.

Inputs that matter

prompt defaults to empty - yes, that's intentional, the node doesn't force a starter prompt. image and image_url are the reference inputs: connect an IMAGE tensor, or give it a public URL (ignored if the image input is connected). duration runs 5 to 30 seconds; leave it at 5 for tests. resolution is 720p or 480p (default 720p). image_way chooses base64 (default) or image_url for how the reference travels. poll_interval (6s) times max_poll_attempts (600) adds up to a 3600-second ceiling - plenty for any Grok render. api_key is optional once the pack's API Settings node is set.

Install and gotchas

Same pack install as always: ComfyUI Manager → search Comfyui-zhenzhen, or git clone https://github.com/T8mars/Comfyui-zhenzhen into ComfyUI/custom_nodes, then restart. No local models.

The FAL route bills differently from the direct route, and that surprises people: expect the pre-charge/hold behavior rather than a clean per-render price, with the final amount adjusted by actual resolution, duration, and reference count. If a render times out at the Await node, the media is usually finished - the FAL task page / Zhenzhen task page has it. 500 errors are upstream Grok load; the pack's answer is re-run it. And remember the seed input is cache-control only (Grok's API has no native seed), so a fixed seed just replays the cached result rather than giving you reproducibility in the diffusion sense.