zhenzhen-video-gk-v15-lowprice

The long-clip budget seat in the zhenzhen pack

T8mars's whole pitch is "the newest closed video models, priced near cost." This node is the one you reach for when you want length on a budget: zhenzhen-video-gk-v15 over the domestic (国内版) API, 6 to 30 seconds, 480p or 720p. It's not the prettiest model in the pack - that's the omni-flash sibling - but nothing else here lets you stretch a clip past 10 seconds without jumping to 4K pricing. Storyboarding, bulk b-roll, quick draft clips for a longer edit: this is the cheap seat, and it knows it.

How it works

Like every node in the "Seedance2 Low Price" category, this is an HTTP client, not a local model. There are no weights to download - the model lives on a server you can't run at home, because there are no open weights for it at all (closed-source-models.md). You type a prompt, optionally wire in up to 7 reference images, and the node:

1. uploads the reference images to zhenzhen's server
2. submits a generation task
3. polls until the video is ready
4. downloads the MP4 and hands you a ComfyUI VIDEO object

From the canvas it looks like a sampler that happened to return a video. Underneath it's requests doing submit-then-poll against the Seedance-compatible API.

The inputs that actually matter

• prompt - your video description. English serves these models best.
• seconds - 6 through 30, the thing this node exists for.
• resolution - 480p or 720p. 720p is the quality ceiling here; don't expect more.
• ratio - 16:9, 9:16, 1:1, 3:2, 2:3.
• image1…image7 - optional reference images for consistency, which is where the GK class models shine.

One thing you don't set: the API key lives in a separate Seedance 2.0 Low Price Settings node (ZHENZHEN_SEEDANCE2_CONFIG). Add one, paste your base_url and api_key from the domestic site (api.seedance.nz), and wire it into this node's api_config input. No settings node and no SEEDANCE_API_KEY env var means the node refuses to run. The seed input is a red herring in the good way - the tooltip says it's only for ComfyUI cache control, and it is not sent to the API. Change it to randomize to force a fresh run; leave it fixed to reuse the cached result.

Outputs

Four of them: video (the finished clip, ready to wire into a preview or Save Video), video_url (the raw link if you want to grab it yourself), task_id (your recovery handle - if ComfyUI times out, the video is usually done and waiting in the site's async-task page under this ID), and response (the full JSON status, useful for debugging).

Install

The README is one long Chinese changelog, so the install is the boring standard bit:

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

then restart ComfyUI and let it install requirements.txt (mostly packages ComfyUI already has; openai, transformers and opencv-python are the heavier additions). ComfyUI Manager works too - search "Comfyui-zhenzhen".

Gotchas

When you hit a 500, the README's advice is blunt: it's usually the upstream provider, not you - just rerun. If you see SSL errors, the pack bundles its own CA cert for the default endpoint; that's by design, so don't go "fixing" it. And if you're outside China, remember these low-price nodes point at the domestic site by default - connection failures are often region, not code.