Zhenzhen Video GK v1.5

Zhenzhen_Video_GK_V15 is the v1.5 house-brand video node from T8mars/ComfyUI_Seedance: one model (zhenzhen-video-gk-v15), a prompt, optional reference images, and a clip. It's the quiet middle of the Zhenzhen video family - less flexible than the Omni Flash Lowprice node with its four modes, less opinionated about length than the fixed-8-second V3.1. If you want a straightforward text-to-video or a single/first-frame-plus-prompt job on the author's own hosted model, this is the one with the least ceremony.

"House brand" is doing real work here. Unlike the Wan, Vidu, and Nano Banana nodes, which wrap identifiable third-party models, the Zhenzhen video models are served by the author's reseller API and their upstream is opaque - you're paying per clip for a hosted black box whose lineage you can't inspect. The node itself behaves exactly like the pack's other video nodes: upload, submit to /v1/videos, poll, download, decode. There's no local equivalent to point you to because there's nothing open under this name.

How it works

Fill in prompt (up to 20480 characters - and in the multimodal modes you can reference materials in the prompt with @Image 1-style placeholders, though with only two image slots the practical use is limited). seconds runs from 6 up (the dropdown has 25 entries), resolution is 720p or 1080p, and ratio is the standard seven (adaptive through 21:9). negative_prompt is forwarded to the metadata for steering, and seed is seed-native: -1 means random, non-negative values go to metadata.seed.

Reference images image1 and image2 are optional and, when connected, are submitted as images[]. That's the whole interface - a clean, boring, mid-range video node.

Inputs and outputs that matter

• prompt - the creative input; multimodal placeholder syntax supported.
• seconds - from 6 up (25 choices).
• resolution - 720p or 1080p.
• ratio - 7 options, forwarded as metadata.ratio.
• negative_prompt - optional, forwarded to metadata.
• seed - -1 random or forwarded to metadata.seed.
• image1 / image2 - optional references, submitted as images[].

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

The house-brand video models are the least documented part of the pack, so the honest advice is: verify before you commit. Start at 6 seconds and 720p, confirm the result decodes cleanly, and only then scale duration and resolution - the pack's own guidance across every video node. Keep skip_error in mind for batches (on failure it returns a placeholder error video instead of killing the queue), and remember the per-clip economics plus the closed-model moderation. If you compare this against the pack's other video nodes and can't tell the quality apart, that's not you being dense - the differences are in the hosted models, and only your own test clips will reveal which one you actually like.