Concurrent Submit | Zhenzhen MiniMax Hailuo Video

Before H3 came out open-weights, "MiniMax video" meant Hailuo - the closed, hosted video model - and it's still the one most people mean when they say "Hailuo." This node is the Concurrent Submit twin of Comfly_MiniMax_video, the pack's wrapper for Hailuo's API. It's the queue-friendly version: instead of blocking your graph on one Hailuo render, it drops each job into the shared 10-worker video pool and returns a task handle immediately. If you're generating a set of shots for a sequence, that's the whole workflow.

The inputs that matter

• model - Hailuo's model lineup; the author's changelog notes it supports "all Hailuo video models" and points at the official MiniMax docs for the parameter details.
• prompt - required.
• duration - clip length.
• resolution - output size.
• first_frame_image / last_frame_image - image-to-video with a proper start and end frame. This is the "tell me where it starts and where it ends" control that makes transitions look intentional.
• subject_reference - keep a subject consistent across clips, which pairs beautifully with the concurrent setup: same reference, N shots, one collector.
• prompt_optimizer - let the API polish/enhance your prompt before generation.
• fast_pretreatment - a speed-over-polish toggle for draft iterations.
• api_key - direct key paste (main API route).
• seed / skip_error - the standard pack pair.

The concurrent flow

Output is a task (COMFLY_VIDEO_FUTURE), never a video directly. The pattern is the pack's usual one: several Submit nodes (same subject reference, different prompts or durations) → every task into ComflyConcurrent_Video_Await (10 slots) → collector returns the finished clips in slot order plus a status JSON. COMFLY_VIDEO_PENDING can add queue depth beyond the 10 workers.

Install and gotchas

Ships in T8mars/Comfyui-zhenzhen - ComfyUI Manager → "Comfyui-zhenzhen", or git clone https://github.com/T8mars/Comfyui-zhenzhen into custom_nodes, then restart. No local models, standard deps. Paid token, overseas servers (proxy from China). Two things the changelog itself warns about: the exact model capabilities live in the official MiniMax docs (read them before assuming a parameter works), and Hailuo's price depends heavily on duration and resolution - a batch of long clips is a batch of long bills.