并发提交｜Zhenzhen Video G Omni Flash

If you read "Omni Flash" as "the fast, cheap one," you've got the right read. Zhenzhen Video G Omni Flash is the shop's quick-video tier - the G line's speed-first member - and this is its concurrent submit wrapper: one model, zhenzhen-video-g-omni-flash, tuned for fast, budget-friendly clips. The naming is marketing, but the positioning is honest: this is the node you reach for when you need a batch of draft clips to pick from, not a hero render.

The inputs are the standard Zhenzhen video set:

• prompt - up to 20,480 characters.
• seconds - 4 is the default (the fastest tier of the pack's video durations).
• resolution - 720p default.
• ratio - 16:9 default, forwarded as metadata.ratio.
• negative_prompt - optional, forwarded to metadata.
• seed - -1 = random; non-negative values forwarded to metadata.seed.
• image1 / image2 - optional reference images, submitted as images[].

Output is the single future (SEEDANCE_VIDEO_FUTURE) into 并发接收视频（10 路）.

How it works

Identical machinery to the other video submit nodes: preflight validation, hand-off to the video thread pool (10 workers; SEEDANCE_VIDEO_CONCURRENCY to adjust), future returned. The request hits the /v1/videos endpoint with the metadata fields above; the receiver waits on all slots, returns videos in order, and handles failures via per-slot skip_error and its own failure_mode. Same shape as the GK v1.5 submit node, different price point and speed.

Installing it

ComfyUI Manager → search ComfyUI Seedance, or:

cd ComfyUI/custom_nodes
git clone https://github.com/T8mars/ComfyUI_Seedance.git
python -m pip install -r ComfyUI_Seedance/requirements.txt

Restart, look under Seedance/并发提交. requests and an api.seedance.nz key; nothing to download.

Where people get burned

• It's the cheap tier, so it looks like the cheap tier. "Omni Flash" doesn't mean 4k; it means fast drafts. Expect fast, expect budget, don't expect the flagship to be hiding inside the price point.
• Duration and resolution discipline. The defaults (4 seconds, 720p) are the sane starting point; every step up in either dimension multiplies render cost across ten parallel slots.
• Seed forwarding. With -1 across the board you get ten random clips you can't diff. For a batch whose whole point is comparison, fixed seeds per slot are your friend.
• future is not a video. Receiver node, then SaveVideo.

Standing fine print: metered per clip, expiring result links, data leaves the machine. When you want ten fast drafts of a storyboard beat and you don't want to think about VRAM, this node plus the 10-路 receiver is exactly the right amount of node.