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

The pack's flagship image node, put on the conveyor belt. zhenzhen-image-g-v2-lowprice is the one that goes to 4k, takes up to sixteen reference images, and covers both text-to-image and reference-image editing. Its Concurrent Submit wrapper runs all of that through the shared 30-slot image pool. For batch product work, character consistency tests, or any job where you want a grid of variations at high resolution, this is the node most of the pack's users reach for first.

G-V2 is the "big one" in the pack's image lineup: 1k/2k/4k resolution, 12 aspect ratios including the ultrawide 21:9 / 9:21, a custom size escape hatch, and up to 16 reference images for multi-image edits. It's a closed Google-family model behind a reseller API - you can't run it locally, period. That's the whole reason this node exists.

How the wrapper works

Standard concurrent machinery, executed well: the node mirrors the original's inputs, validates, and drops the job into a bounded thread pool - 30 image workers, 10 video, both tunable via COMFLY_IMAGE_CONCURRENCY / COMFLY_VIDEO_CONCURRENCY. The task output (COMFLY_IMAGE_FUTURE) goes into Concurrent Collect Images (30), which waits for all your submissions and returns IMAGE tensors in connection order plus a status summary. Sixteen reference images on one node is heavy; the pool is what lets you still queue a dozen of those nodes at once without your machine caring.

Inputs that matter

• model - fixed to zhenzhen-image-g-v2-lowprice. No dropdown drama here.
• prompt - the instruction. Multiline.
• resolution - 1k, 2k, 4k. This is the main cost lever: 4k bills more than 1k, always.
• size - twelve options: 1:1, 16:9, 9:16, 21:9, 9:21, 4:3, 3:4, 3:2, 2:3, 4:5, 5:4, and custom. The README notes the dropdown was reworked so old workflows that saved a raw WxH auto-migrate.
• custom_size - appears when size is custom; format 1024x1024. The README's old advice that custom sizes need to divide by 16 still applies in spirit - this is a cloud model with its own resolution rules.
• n - 1–10 images per call. The billing multiplier.
• image1–image16 - reference images. Text-to-image with none connected, multi-reference edit with a pile of them.
• api_config - zhenzhen-seedance2-low-price-api-settings, domestic key from api.seedance.nz.
• seed - cache-control only; never sent to APIs without a native seed.

Install

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

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

No model files. An API key is the whole install.

Where people get burned

• task → collector, never → Save Image. The universal family trap, and it stings more here because a 4k batch failure feels expensive.
• 16 references × 30 slots is 480 uploaded images. Reference-image uploads are billed extra in this pack (the README is explicit for the FAL nodes, and the image nodes follow the same upload model). Big multi-reference batches are the fastest way to surprise yourself with a bill.
• 4k is slow and pricey. The pool makes it parallel, not free. If you're batch-testing prompts, stay at 1k until you know what you want, then send the winners through at 4k. That two-stage habit is the single biggest saver with this pack.
• Custom size has rules. The WxH format is strict (widthxheight, no spaces), and the model will bounce sizes it can't handle. Read the node's error before assuming the API is down.

Reach for G-V2 when you need resolution and multi-reference muscle. The 4k flag is the headline, but the sixteen-reference edit is the sleeper feature - consistency across a whole product line, in one queued batch.