Concurrent Submit | Zhenzhen_vidu_text2video

Text-to-video is the lazy Vidu lane: no reference image to hunt for, just a prompt and a clip comes back. It's also the lane where you want to generate a batch - the model's hit rate on "does this match what I imagined" is middling, so the practical workflow is six prompts, pick the one that lands. That's exactly what this concurrent wrapper is built for. It submits to the pack's shared video pool (ten workers) and returns a task immediately, so you can fire off a whole idea board of Vidu text-to-video drafts at once and collect them in the Concurrent Collect Videos node.

Inputs that matter

• prompt - required, and it's the whole show here. Vidu reads natural-language motion descriptions reasonably well; "camera slowly pushes in while the character turns" works better than tag spam.
• model - viduq2 (default), viduq1, or vidu1.5. Q2 is the current generation; the older ones are cheaper/faster if you're iterating.
• style - general or anime. The anime lane is a genuine different model behavior, not a filter.
• duration - 1 to 10 seconds, default 5.
• aspect_ratio - 16:9, 9:16, 3:4, 4:3, or 1:1.
• resolution - 360p, 540p, 720p (default), or 1080p.
• movement_amplitude - auto, small, medium, large. Same motion dial as the rest of the Vidu family.
• bgm, off_peak, watermark, wm_position - background music, cheaper off-peak billing, watermark placement.
• api_key, seed, skip_error - standard.

Single output: a COMFLY_VIDEO_FUTURE task.

How it works

Standard Zhenzhen concurrent machinery: the wrapper in ComflyConcurrent.py validates, then hands the real Comfly_vidu_text2video call to a bounded executor with ten video workers and returns the future. The collector (ComflyConcurrent_Video_Await) waits on every connected task, restores slot order, and reports per-slot status. failure_mode: placeholder swaps a failed slot for a blank clip instead of aborting the batch - useful when you're intentionally generating ten and only care about the survivors.

Setup

Install the pack once - ComfyUI Manager (search "Comfyui-zhenzhen"), or:

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

No models, no VRAM. You need a paid Zhenzhen key from ai.t8star.org (overseas) or api.seedance.nz (domestic). Every clip is a metered call, so a ten-wide idea board is ten charges - cheap at 540p, not so cheap at 1080p with music.

Gotchas

Text-to-video without a reference frame is where Vidu's prompt adherence is weakest, so don't be surprised when draft four of ten is a different shot than draft two - that's the model, and it's the argument for the concurrent version in the first place. movement_amplitude: large is tempting and frequently looks worse; auto is a better default for anything you'll actually use. And the ten-slot pool is shared across every video node in the pack, so if your graph already has Sora and Veo jobs queued, your Vidu drafts wait their turn.