Concurrent Submit | zhenzhen-veo-omni

Google's flash-tier video models are the cheap door into Veo, and omni_flash-10s is the one this node wraps. The pack's changelog pins the price at 1 point per 10-second clip, which makes it the budget Veo - the lane you use when you want Google's video quality for draft-and-iterate work without Veo 3.1 Pro pricing. It also handles both image-to-video and video-editing mode, so one node covers "animate this still" and "restyle this clip."

The concurrent wrapper is the standard Zhenzhen Submit pattern. It doesn't block; it submits to the shared video pool (ten workers) and hands you a task immediately. Wire it into the Concurrent Collect Videos node and you can run a dozen cheap Veo drafts in parallel - which, at one credit a pop, is the sensible way to use this node.

Inputs that matter

• prompt - defaults to the Chinese phrase "基于这张图片生成视频" ("generate a video based on this image"). Write your own; it's just a starting default.
• mode - img2video (default) or video_edit. The mode decides which reference sockets matter.
• model - omni_flash-10s, the only choice.
• size - 1280x720 or 720x1280. No 4k here, and that's the point.
• seconds - 4, 5, 6, 8, or 10.
• image / image_url - reference for img2video. A connected image wins over the URL.
• video / video_url - reference for video_edit mode. video_way (upload vs video_url) controls how your reference video gets to the API - pick upload when you're feeding a connected IO.VIDEO, URL when you already have one hosted.
• watermark - off by default; turn it on if you want Google's watermark for whatever reason.
• apikey, poll_interval, max_poll_attempts, seed, skip_error - standard.

Output: a single COMFLY_VIDEO_FUTURE task.

How it works

Same machinery as every Submit node in the pack: the ComflyConcurrent.py wrapper copies the original Comfly_veo_omini inputs, runs its validator, and drops the real generation call onto a bounded executor with ten video workers. The returned task is a future the collector waits on; the collector restores your slot order and reports per-slot status. Set failure_mode to placeholder and a failed slot yields an empty video rather than a crashed batch.

Setup

One install covers all of it - ComfyUI Manager (search "Comfyui-zhenzhen"), or:

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

No models to download; every clip is a paid cloud call. You need a Zhenzhen key from ai.t8star.org (overseas) or api.seedance.nz (domestic). If the "1 point / 10s" price sounds great, it is - just don't confuse this flash lane with Veo 3.1's quality tier.

Gotchas

At one credit a clip, this is the node to reach for when you want volume - but the flash tier's output is visibly lighter than the pro Veo models, so don't batch thirty of these expecting cinematic 4k. If a video_edit run misbehaves, check which video_way you're on; a URL that's expired or unreachable from the server is the classic silent failure. And as always with this pack: timeouts aren't guaranteed failures - check your async task page - and the overseas endpoint needs a working proxy from many regions.