Zhenzhen Video G Omni 1.1 Flash Lowprice（4 模式）

The name is a mouthful, so here's the short version: this node sends a prompt to a cloud video model called zhenzhen-video-g-omni-1.1-flash-lowprice and drops a finished video back into your graph. No local weights, no VRAM, no ControlNet spaghetti. If your GPU is a potato or you just don't feel like babysitting a 10-minute local render, this is the "video from a text box" experience - with a price tag.

It's one node in ComfyUI_Seedance (T8mars), a reseller-style API pack that fronts api.seedance.nz (domestic) and ai.t8star.org (overseas). The pack is a buffet: Seedance 2.0/2.5, Kling, Hailuo, Vidu, Wan, plus image/audio/music/3D nodes, all behind the same key. The "Zhenzhen Video G Omni 1.1 Flash" model is the shop's own naming on its "Lowprice" tier - you won't find this ID in any official ByteDance docs, and "Zhenzhen Video" has essentially zero community footprint. It's a long-tail reseller product, which is exactly what the KB's API-wrapper notes warn you to approach with open eyes: this category ships a key and phones home by design.

What the mode switch actually does

The whole point of this node is the mode dropdown, and it's stricter than you'd expect. Pick one:

• text - prompt only. Attach any media and it refuses before uploading.
• frame - exactly one first-frame image in image1.
• reference_images - a contiguous run of 1 or 3 images (image1, or image1..image3).
• reference_video - exactly one reference clip, either a local input_video or a public video_url, never both.

The validation runs before anything is uploaded, so a wrong combo fails fast with a clear message instead of burning an API call. That's a genuinely thoughtful touch for a paid service. In reference_video mode the node also quietly drops seconds from the request - the duration is taken from your clip.

The inputs that matter

The rest is a short list: prompt (up to 20480 chars), seconds (4/6/8/10, default 6), resolution (720p/1080p/4k, default 720p), aspect_ratio (16:9 or 9:16), and nsfw_check to flip on the documented content filter. Note the pack's docs treat 1080p/4k as upscaled output tiers rather than native renders - start at 720p.

skip_error returns a placeholder video instead of killing the workflow when a call fails - handy in a batch. And seed here is purely a ComfyUI cache key: set it fixed and unchanged inputs reuse the cached result; the value isn't sent to this model. Wire api_config from the Seedance API Config node, or skip it and rely on the SEEDANCE_API_KEY env var / config/.env.

Outputs

Four of them: video (the downloaded local clip - that's the one you connect to SaveVideo or a preview node), video_url (the direct link), task_id (handy if you ever need to chase the shop about a job), and response (the raw JSON, for the curious).

Install and the catch

It's the standard pack install: ComfyUI Manager → search ComfyUI Seedance, or

cd ComfyUI/custom_nodes
git clone https://github.com/T8mars/ComfyUI_Seedance.git

The only Python dependency is requests - there are no model files to download because there's nothing local. Grab an API token from the seedance.nz console and you're set.

Where people get burned: every run is metered, and video minutes add up fast - this category is the community's "right tool for models you can't run locally, wrong default for ones you can" line. Test with 4s/720p before you commit to 4k. And since this is a third-party pack holding your key and calling the network as its whole job, give it a read before you trust it with real work.