Concurrent Submit | zhenzhen-hailuo-h3-video-lowprice

Hailuo H3 is MiniMax's current generation, and this node is the fullest expression of it in the pack: not just text-to-video and image-to-video, but multi and global variants, plus an input list that goes well past one reference image - up to nine images, three videos, and three audio clips in a single call. That's the "seedance-style" multi-reference pattern the pack keeps building on, and it's how you get consistent characters or reference audio driving a scene.

It's also a low-price-family node, so it talks to the domestic api.seedance.nz storefront at the discounted rate, and it's a concurrent submit node, so ten of these multi-reference calls can run at once in the shared video pool. If you want character consistency on a budget, this is one of the more serious options in the pack.

How the wrapper works

The plain Comfly_hailuo_h3_video_lowprice node is a Seedance2 Low Price unified node: it takes an api_config (ZHENZHEN_SEEDANCE2_CONFIG from the pack's settings node) for the domestic URL and key, assembles the multi-media references, submits, polls, and downloads. The Submit wrapper keeps every input but returns a single task (COMFLY_VIDEO_FUTURE), running the call on the video worker pool (default 10, COMFLY_VIDEO_CONCURRENCY env var).

The task must go into ComflyConcurrent_Video_Await, which waits for all ten slots and returns video_1 … video_10 in submission order plus a status JSON. failure_mode (fail_fast / placeholder) controls whether one bad call aborts the batch or blanks the slot.

Inputs that matter

model picks the mode: hailuo-h3-t2v, hailuo-h3-i2v, hailuo-h3-multi, hailuo-h3-global-t2v, hailuo-h3-global-i2v, hailuo-h3-global-multi - the global variants are the wide-moderation/looser-filtering tier the pack advertises for e-commerce and global-market content. prompt is required; seconds runs 5–15; resolution is 768P or 2K (note: H3 goes to 2K, a step above Hailuo 2.3's 1080p); ratio has the full set (adaptive, 16:9, 4:3, 1:1, 3:4, 9:16, 21:9).

The optional inputs are where H3 shows off: image1–image9 (IMAGE), video1–video3 (VIDEO), and audio1–audio3 (AUDIO) all feed the multi and global-multi models. Reference audio plus reference video in one generation is the thing you can't get from most other nodes in this pack. api_config connects the domestic settings node; seed is cache-control only.

Install and gotchas

Pack install as always: ComfyUI Manager → search Comfyui-zhenzhen, or git clone https://github.com/T8mars/Comfyui-zhenzhen into ComfyUI/custom_nodes, restart. No model downloads.

The low-price family gotcha applies fully here: this node needs the domestic api.seedance.nz config and its own key - the overseas ai.t8star.org key won't authenticate. The multi modes are also heavier on both token cost and render time than plain t2v; the pack's own guidance is to use global variants deliberately, not as a default, because the reference count directly moves the price. If a big multi-reference render times out at the Await node, check the site's task page before re-running - the clip is probably done and the poll just expired.