Concurrent Submit | zhenzhen-video-g-omni-flash-lowprice

The modern omni-flash video node, and the one you should be using. zhenzhen-video-g-omni-flash-lowprice (v2 - the "旧版兼容" legacy node is a different, older thing) is Google's fast video model with an explicit four-way mode switch, resolution up to 4k, and durations of 4/6/8/10 seconds. Its Concurrent Submit wrapper queues it through the pack's 10-slot video pool.

omni-flash is Google's speed-first video tier - the README's earliest note on it (as omni_flash-10s) priced it at one credit per 10 seconds and described it as the fast draft option. The v2 node is the author's cleanup of that: instead of guessing what you want from what you connect, you pick the mode and the node shows you the relevant inputs.

The four modes

• text - pure text-to-video. Prompt in, clip out.
• frame - image-to-video from a first frame (the image1 input).
• reference_images - a video influenced by reference images (image1–image3), the model-consistency path.
• reference_video - video influenced by another video (input_video / video_url).

The explicit mode dropdown is the whole quality-of-life win over the legacy node: fewer ambiguous builds, clearer validation, less "why did it ignore my reference?" guessing.

Inputs that matter

• mode - text / frame / reference_images / reference_video. Pick first; it decides which optional inputs matter.
• prompt - the scene description. Multiline.
• seconds - 4, 6, 8, 10. Longer is slower and pricier; 4–6 seconds is the draft sweet spot.
• resolution - 720p, 1080p, 4k. This is Google's fast tier - 4k here is "fast-tier 4k," not premium-model 4k, so manage expectations. It's also the main cost dial.
• aspect_ratio - 16:9 or 9:16. Short and simple, which tells you the target use is vertical short-form.
• nsfw_check - a boolean sent to the API. Default off.
• image1–image3 / input_video / video_url - the reference media, used depending on mode.
• api_config - zhenzhen-seedance2-low-price-api-settings (domestic key from api.seedance.nz).
• seed - cache-control only.

Output: task (COMFLY_VIDEO_FUTURE) → Concurrent Collect Videos (10) → VIDEO tensors + status.

Install

ComfyUI Manager → search "Comfyui-zhenzhen" → install → restart, or:

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

No model files. API key required.

Where people get burned

• Mode still wins over connection. Setting mode to reference_images but only wiring input_video is a contradiction the API has to resolve - read the mode you picked, wire accordingly.
• "4k" is a tier label, not a promise. omni-flash is the cheap/fast tier; if you need premium 4k video, that's a different (pricier) node in the pack. Draft in 720p, commit to 4k only for the keepers.
• The video pool is ten. Ten slots, and video jobs are long. A full queue means a long wait - queue consciously.
• Timeout ≠ failure. When a video slot errors, check the site's async task list before re-billing; your clip may have finished server-side.

This is the video node to reach for when you want Google's fast tier with clear controls. Mode dropdown, reasonable durations, up to 4k when you need it - and a 10-wide queue so you can shoot a batch of drafts in parallel without melting your bill.