Zhenzhen Video G Omni Flash Lowprice（4 模式）

If Zhenzhen_Video_G_Omni_Flash is the plain one, this is its overachieving sibling. Zhenzhen_Video_G_Omni_Flash_Lowprice is the same "Omni Flash" model family but with a mode dropdown that turns it into four nodes at once - text, frame, reference_images, and reference_video - and it's the only video node in T8mars/ComfyUI_Seedance that offers 4k resolution. The pack's changelog even calls it a real engineering effort: verified end-to-end across all four modes, including reference-video work.

"Lowprice" is the brand's budget tier, but don't read that as weak - read it as "the tier where they bundle the most modes." If you're going to use one Zhenzhen video node, this is the one to try first, because the mode switcher lets you compare text-to-video against first-frame, multi-reference, and reference-video on a single node before you commit to any of the fixed models elsewhere in the pack.

How it works

The mode value decides both the inputs and the request shape:

• text - pure text-to-video; prompt plus seconds (4/6/8/10), no media.
• frame - first-frame animation; image1 drives it.
• reference_images - style/character reference; accepts 1 or 3 images (image1–image3).
• reference_video - video-to-video; feed input_video (local, auto-uploaded) or a public video_url, and seconds is automatically omitted because the model follows the source.

Across modes: resolution is 720p/1080p/4k, aspect_ratio is 16:9 or 9:16, and there's an nsfw_check toggle that enables the documented content check (default off - you opt in to stricter filtering). As with every node here, it's submit → poll → download → decode, with the pack's defensive download chain underneath.

Inputs and outputs that matter

• mode - text / frame / reference_images / reference_video. The whole node.
• prompt - the creative input; supports @Image 1-style references in multimodal modes.
• seconds - 4/6/8/10; auto-omitted in reference_video mode.
• resolution - 720p, 1080p, or 4k.
• aspect_ratio - 16:9 or 9:16.
• nsfw_check - content-check toggle, default off.
• image1–image3, input_video, video_url - media per mode.

Outputs: video, video_url, task_id, response.

Install

One node in T8mars/ComfyUI_Seedance:

# ComfyUI Manager: search "ComfyUI Seedance", or:
comfy node install seedance

Or clone https://github.com/T8mars/ComfyUI_Seedance.git into ComfyUI/custom_nodes, pip install -r requirements.txt, restart, and configure a key from api.seedance.nz/console via the Seedance API Config node or SEEDANCE_API_KEY.

Where people get burned

Three spots. First, the mode-driven UI hides irrelevant inputs, but a stale connection can outlive a mode switch - if you switch from reference_images to text and leave an image wired, check what's actually in the request (the response output shows it). Second, reference_video mode ignores seconds, so a 10-second source stays 10 seconds no matter what the dropdown says. Third, 4k is 4k: it's the most expensive clip this pack produces, and the default download cap is 8 GiB (SEEDANCE_VIDEO_MAX_MIB), so keep the source short. And the house-brand caveat holds - the upstream is opaque, per-clip metered, and moderation applies to your prompt no matter which mode you pick.