Zhenzhen Video V3.1

Zhenzhen_Video_V31 is the flagship-length node of the house-brand video family in T8mars/ComfyUI_Seedance. Its signature move is refusing to negotiate on duration: seconds is fixed at 8, period. What it gives you instead is a three-tier model choice - zhenzhen-video-v31-fast, zhenzhen-video-v31-quality, and zhenzhen-video-v31-lite - plus a 4k resolution option that only the Lowprice Omni Flash node shares. Fast/quality/lite is the classic pricing ladder: iterate on fast, present with quality, and use lite when you just need a quick clip with no references at all.

That fixed 8 seconds is worth a second of thought. Most pack nodes let you stretch to 15 or 30 seconds; V3.1 pins you at a social-clip length, which tells you what this model was built for - finished 8-second pieces for feeds, not long takes. If that fits your use case, this is arguably the most "presentable output" node in the Zhenzhen line: three tiers, two ratios, up to 4k.

How it works

The model choice is the whole game:

• zhenzhen-video-v31-fast - the quick lane; supports up to 3 reference images (image1–image3).
• zhenzhen-video-v31-quality - the quality lane; up to 2 reference images.
• zhenzhen-video-v31-lite - text-to-video only; no image inputs at all (the node hides them).

Everything else is shared: prompt (up to 20480 chars, @Image 1 reference syntax supported where images exist), resolution (720p/1080p/4k), ratio (16:9 or 9:16), negative_prompt forwarded to metadata, and a native seed (-1 random, non-negative forwarded to metadata.seed). Standard async submit → poll → download → decode on the /v1/videos endpoint.

Inputs and outputs that matter

• model - fast / quality / lite, with their reference-image limits.
• prompt - the creative input.
• seconds - fixed at 8.
• resolution - 720p, 1080p, or 4k.
• ratio - 16:9 or 9:16.
• negative_prompt - optional, forwarded to metadata.
• seed - -1 random or forwarded to metadata.seed.
• image1–image3 - per-model limits (fast 3, quality 2, lite 0).

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 model-tier trap is the big one: Quality forbids three-image reference work and Lite forbids all images, so a workflow that works on Fast silently does nothing on Lite if you forget to unwire the references - the node hides the slots, but the request shape still follows the model. Also internalize that 8 seconds is not a bug; if you need 4 or 12, this is the wrong node (try the Omni Flash or GK v1.5 nodes). And as always with the house-brand models: the upstream is an opaque hosted black box, every clip is metered with 4k at the top of the price ladder, and upstream moderation applies regardless. Smoke-test on fast at 720p, then step up to quality at your target resolution once the result reads right.