Zhenzhen Image GK v2 文生图

Zhenzhen_Image_GK_V2 is the plainest node in the pack's house-brand image family: text-to-image and nothing else. Three inputs - prompt, size, n - and it refuses to pretend there's more to it. That's refreshing in a pack where half the nodes are 8-in-1 mashups with dynamic UI. If you want a still generated on your existing key, this is the fastest node to configure in the whole Zhenzhen lineup.

It's worth being clear about what "GK v2" is: the author's house-brand image model, served through the api.seedance.nz reseller that backs the entire T8mars/ComfyUI_Seedance pack. There's no open model to compare it against, no weights to inspect, no local equivalent to name - it's a hosted black box that costs you a per-image call. The GK v2 family does have genuinely interesting siblings (a segmentation node and a region-edit node that operate server-side on GK v2 generations - see Zhenzhen_Image_GK_V2_Segment and Zhenzhen_Image_GK_V2_Region_Edit), and this text-to-image node is the front door to that whole chain.

How it works

Prompt goes up, task gets submitted to the reseller's /v1/image/generations endpoint, the node polls until it's done, downloads the result, and hands you a decoded ComfyUI IMAGE. No reference images, no negative prompt, no thinking-mode toggle - if you want editing or a mask, the v2 family routes you through the dedicated edit/segment/region nodes instead of cramming it in here. The size dropdown offers seven fixed ratios (1:1, 2:3, 3:2, 3:4, 4:3, 9:16, 16:9), and n runs 1 to 12.

Same cost note as its siblings: n above 1 asks the API for that many images, but the node downloads only the primary result - the extras live in the response JSON and your bill. There's a 30-image concurrent submitter for this family elsewhere in the pack if you genuinely need volume.

Inputs and outputs that matter

• prompt - up to 20000 characters.
• size - seven fixed aspect ratios.
• n - 1 to 12; downloads the primary result.

Outputs: image, image_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 add a key from api.seedance.nz/console via the Seedance API Config node or SEEDANCE_API_KEY.

Where people get burned

The biggest trap is the n meter: set it to 12 out of curiosity and you've paid for twelve images and kept one. Everything else is the pack's standard advice - the prompt is the whole creative surface here, so spend it well, remember the hosted model enforces its own moderation (no bypassing from a node), and keep skip_error on if you're batching so one rejected prompt doesn't stop the line. And if you find yourself reaching for this node repeatedly for the same job, ask whether a local open model would do the still generation for free while you reserve the paid path for the closed-model features.