Concurrent Submit | zhenzhen-video-gk-v15-lowprice

Text-to-video and image-to-video for the GK V1.5 tier, in a pool of ten. zhenzhen-video-gk-v15-lowprice takes a prompt - plus up to seven reference images for the image-to-video side - and returns a video clip. Its Concurrent Submit wrapper queues those jobs through the pack's video pool, which is capped at 10 workers, not 30.

Video is where the "low price" pitch gets real, because video APIs are the expensive category. GK V1.5 video is one of the budget tiers of the pack - 480p and 720p only, which tells you what it's for: fast drafts, storyboard tests, short-form content slop-checks before you spend on the premium video models. At 480p with a 6-second clip, this is the cheapest way to see if a scene idea holds.

How it works

The familiar machinery, video edition: Submit nodes mirror the original's inputs, validate, and hand jobs to a bounded ThreadPoolExecutor (10 video workers, tunable via COMFLY_VIDEO_CONCURRENCY). Each worker does the submit → poll → download cycle on its own thread. Output is a task of type COMFLY_VIDEO_FUTURE, which feeds into Concurrent Collect Videos (10) - it waits for all ten slots and returns real VIDEO tensors in connection order, plus a status string.

The 10-slot video cap is the whole design: video jobs are long and expensive, and the author chose ten workers rather than thirty. If you queue more than ten, the rest wait for a slot. That's a feature, not a bug - it's the pack's way of not letting you explode your bill in one click.

Inputs that matter

• prompt - the scene description. Multiline. For reference-image mode, this does the talking about how the refs should move.
• seconds - a long enum from 6 up to ~30. Longer = more expensive. For drafts, 6 seconds is the sensible default.
• resolution - 480p or 720p. The budget tier's ceiling. This is how the pack prices this node low.
• ratio - 16:9, 9:16, 1:1, 3:2, 2:3.
• image1–image7 - reference frames/images for the reference-to-video path (the README's V1.5 note says up to 7 refs). Leave empty for pure text-to-video.
• api_config - zhenzhen-seedance2-low-price-api-settings, domestic key from api.seedance.nz.
• seed - cache-control only; not a real video seed.

Install

ComfyUI Manager → search "Comfyui-zhenzhen" → install → restart, or:

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

No downloads. API key required.

Where people get burned

• Video task → Video collector. Pair this with Concurrent Collect Videos (10), not the image one. The type error is immediate and the fix is boring.
• Seconds × resolution × slots is the invoice. Ten slots at 720p/15s is a real bill. The pool makes parallel video feel as cheap as parallel images - it isn't.
• Long jobs time out on the node but not on the server. The README's whole Sora section is a masterclass in this: node timeout ≠ job failure. If a video slot reports failed, go check the site's async-task list - your MP4 may be sitting there, generated and waiting.
• 480p is 480p. Don't queue a 10-slot draft batch and expect deliverables. This tier is for iteration; the upscaler node exists for the upgrade path.

The bottom line: this is the pack's cheap video workhorse. Use it to iterate fast on scenes, keep the resolution honest, and reserve the premium video models for the clips that survive the draft.