zhenzhen-video-g-omni-flash-lowprice

The omni-flash node you should actually add

This is the documented, current version of zhenzhen's Google omni-flash video node, and the thing that sets it apart from its legacy sibling is a single dropdown: mode. Instead of guessing what you meant from which inputs you plugged in, v2 makes you say it - text, frame, reference_images, or reference_video - and then validates that your inputs match the mode you picked. It's the difference between a magic box and a tool with a manual, and for a paid cloud call the manual wins every time.

A quick word on what "omni-flash" is: it's Google's low-price video model, and like every closed model in this pack there are no open weights (closed-source-models.md). You're renting it by the call through zhenzhen's Seedance-compatible API. What you get for the money is a model that does text, image, and video input in one box - hence "omni."

The modes, briefly

• text - text-to-video. The only mode that rejects reference media entirely; connect nothing but the prompt.
• frame - first-frame animation. Requires exactly image1, no more, no fewer.
• reference_images - the character-consistency mode. Either image1 on its own or image1 through image3; no video allowed.
• reference_video - style/continuity from a clip. Exactly one of input_video or video_url, no images.

The validation is strict about these, and it's all good - the error messages tell you exactly which input broke the rule.

The settings that matter

• seconds - 4, 6, 8, or 10. (Note: it's ignored in reference_video mode, which the server drives.)
• resolution - 720p, 1080p, or 4k. That's the real price lever; 4k costs noticeably more.
• aspect_ratio - 16:9 or 9:16.
• nsfw_check - off by default; the upstream filter applies regardless, this just makes the node explicitly request it.
• api_config - the shared Seedance 2.0 Low Price Settings node with your domestic-site key. Same as every node in this category.
• seed - ComfyUI cache control only, not sent to the API, per the tooltip. Randomize to force a new run.

Outputs and install

Standard zhenzhen quartet: video (a real ComfyUI VIDEO you can preview or save directly), video_url, task_id (recovery handle if ComfyUI times out mid-poll), and response (full JSON). Install is pack-wide - ComfyUI Manager, search "Comfyui-zhenzhen", or:

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

then restart. The README's standing advice applies here too: a 500 is usually the upstream provider having a moment - rerun before you blame your prompt. And if a run times out on the node side but you can see the task in the site's async list, that's the same thing: generation finished, download stalled, grab the MP4 from the site.