Concurrent Submit | zhenzhen-hailuo-h3-max-video-lowprice（4合1）

This is the batch twin of the plain Comfly_hailuo_h3_max_video_lowprice node from the T8mars/Comfyui-zhenzhen pack. Same two Hailuo H3 Max models (hailuo-h3-max-t2v and hailuo-h3-max-i2v), same inputs, same cloud API under the hood - but instead of blocking your graph while it uploads, polls and downloads one clip at a time, it throws each job into a shared worker pool and hands you back a task handle almost instantly. You collect up to ten of those handles in one place and wait for the whole batch.

Reach for it when you have a shot list, a set of prompt variants, or a queue of image-to-video jobs you'd rather fire off at once than babysit serially. The individual node's per-clip cost and quality don't change - this is purely a throughput play.

How the pool works

The pack generates a "Concurrent Submit" version of every eligible video node automatically. The wrapper hands your parameters to a bounded thread pool - the video pool defaults to 10 workers, tuneable with the COMFLY_VIDEO_CONCURRENCY environment variable, with a pending queue behind it via COMFLY_VIDEO_PENDING. Instead of a downloaded video, it returns a COMFLY_VIDEO_FUTURE task object. Because the pool is shared across the pack, mixing this node with other zhenzhen video submit nodes just fills the same ten slots.

The task handle goes into the companion collector node, ComfyConcurrent_Video_Await ("Concurrent Collect Videos (10)"), under zhenzhen/Concurrent Collect. Wire up to ten task slots into it and it blocks until every one resolves, then hands back video_1 through video_10 in slot order plus a status string. The UI stays responsive while the pool churns - no more staring at one progress bar for five minutes.

Inputs and outputs

The input side is identical to the plain node, and the same logic applies: pick a model, write a prompt, choose seconds (5–15), resolution (480P/768P only - this is the budget tier) and, for text-to-video, a ratio. In i2v mode image1 is the required first frame and image2 the optional last frame, and the ratio widget hides itself. api_config carries the key from a zhenzhen Settings node. The seed widget is ComfyUI cache control only - fixed means an unchanged slot won't resubmit and re-bill you.

The single output is the task, which doesn't render anything on its own. You must feed it into the collector - that's where the waiting happens and where the real videos come out. One failure_mode choice lives on the collector: fail_fast (default) aborts the batch on the first error, while placeholder swaps in an error video for the dead slot and lets the rest finish.

Installing and finding it

Both nodes ship in the same pack, so install once - ComfyUI Manager, search "Comfyui-zhenzhen", or:

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

Then restart. No models to download; everything runs through zhenzhen's API, so the real setup step is a paid API key from their storefront pasted into the Settings node. Look for the node under zhenzhen/Seedance2 Low Price/Concurrent Submit - search "Concurrent Submit" plus the Hailuo name, or just grab it from one of the pack's example workflows, which already have the Submit nodes wired into the collector.

Gotchas worth knowing

Firing ten jobs at once spends credits in a burst, so be deliberate: every slot is a separate paid cloud call, and a re-executed graph with seeds on randomize bills you again. If one slot's provider call hiccups - the pack's README is honest that 500s happen and reruns fix most of them - fail_fast will sink the whole batch, so set placeholder unless you want all-or-nothing. A timeout doesn't always mean a failed job; check the task on the zhenzhen site console before assuming you lost the money. And because ten videos are downloading at once, slow or VPN-dependent networks (the README recommends TUN mode for reaching the API) will make themselves known quickly.