Concurrent Submit | zhenzhen-image-gk-v2-lowprice

Notice what this one doesn't have: no image inputs. zhenzhen-image-gk-v2-lowprice is the pure text-to-image member of the GK V2 family, and its Concurrent Submit wrapper is the batch version. Same 30-slot shared image pool as the rest of the family, same task-out / collect-later pattern - just no reference images to juggle. If you want a wall of generations from a list of prompts, this is the simplest node in the whole concurrent lineup to set up, because there's nothing to wire except prompts, a config, and a collector.

GK V2's text-to-image is the cheapest and fastest tier of the pack's Gemini-image lineup, and it has the largest per-call batch of any image node in the pack: n goes up to 12. The edit sibling caps at 10, the G-2 at 10 - this one goes to 12. The pack clearly expects you to use that.

How it works

The wrapper is auto-generated by ComflyConcurrent.py: identical inputs to the original node, plus a task output (COMFLY_IMAGE_FUTURE). Submit → job lands in the bounded thread pool (30 image workers) → each worker runs the full submit/poll/download cycle → the Concurrent Collect Images (30) node waits for all your tasks and returns IMAGE tensors in the order you connected them. The status output tells you which slot succeeded, failed, or never got connected.

Inputs that matter

• prompt - that's the whole creative input. Multiline. This is text-to-image with no image conditioning, so the prompt does all the work.
• size - 1:1, 2:3, 3:2, 3:4, 4:3, 9:16, 16:9. Seven shapes, no custom. For a model that otherwise gives you freedom, the ratio list is oddly short - but they're the ones that matter.
• n - 1 to 12 images per call. The billing multiplier. One API call, up to 12 images, but the site charges per image. Twelve at n=12 is twelve charges - worth doing the math on before you queue five Submit nodes.
• api_config - from zhenzhen-seedance2-low-price-api-settings (domestic store key, api.seedance.nz).
• seed - the cache-control seed. Fixed reuses the cached result; randomize forces a new run. Not sent to APIs that lack a native seed, so it can't reproduce the same picture twice.

Install

ComfyUI Manager → search "Comfyui-zhenzhen" → install → restart, or clone into custom_nodes:

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

No downloads, no VRAM, no checkpoints. An API key from the Zhenzhen store is the entire requirement.

The two things to remember

• task is a promise, not a picture. Feed the task outputs into Concurrent Collect Images (30) and save what comes out of there. This is the family's universal trap.
• n is the wallet's lever. The pool makes 30 parallel calls feel free; the invoice disagrees. The whole "lowprice" promise of the pack depends on you not cranking n and batch count simultaneously.

One thing people miss: because there are no image inputs, this node is a great scale test for whether you've got your Zhenzhen API key and config right. One Submit node + one collector, prompt in, images out - the whole concurrency machinery visible in five nodes. Get that working and every other concurrent image node in the pack is the same pattern with extra inputs.