Concurrent Submit | zhenzhen-image-gk-v15-lowprice

zhenzhen-image-gk-v15-lowprice is the pack's merged GK V1.5 model - one node that covers both text-to-image and image editing. This Submit node is the version of it that runs inside the shared 30-slot image pool, so you can queue a whole batch instead of babysitting one generation at a time. If you've got a grid of prompts, a style test, or a pile of reference images that all need the same edit, this is the node you wire up when you don't want to sit there clicking Run.

GK V1.5 (the "gk" is the pack's name for this Gemini-image-family tier) arrives here in two flavors, and unlike the more fiddly GK V2 nodes, the only thing that switches between them is the model dropdown.

How the concurrent wrapper works

This is an auto-generated wrapper from the pack's ComflyConcurrent.py. You get a Submit node with the same inputs as the original, plus a task output of type COMFLY_IMAGE_FUTURE. Behind the scenes it validates your inputs, then drops the job into a bounded thread pool - 30 workers for images, 10 for video. Each worker does the full submit → poll → download dance on its own thread, so ten Submit nodes can genuinely run at once.

The task is a promise, not a result. Wire it into Concurrent Collect Images (30), the pack's collector node, and it hands back real IMAGE tensors in the order you connected them, plus a status JSON describing what each slot did.

Inputs that matter

• model - zhenzhen-image-gk-v15 (text-to-image) or zhenzhen-image-gk-v15-edit (image editing). Pick edit only when you're feeding reference images.
• prompt - the text. Multiline.
• size - the usual aspect ratios: 1:1, 16:9, 9:16, 3:2, 2:3. No custom size here, so pick the nearest shape.
• n - how many images per API call, 1–10. This is where the bill multiplies: n images means n charges, not one.
• image1 - one optional reference image for the edit model. (V1.5's edit path is single-reference; the GK V2 edit node is the one that takes three.)
• api_config - connect the zhenzhen-seedance2-low-price-api-settings node. That's the domestic store's config, key from api.seedance.nz.
• seed - the pack-wide cache-control seed. Fixed reuses the cached result; randomize forces a fresh run. It is not sent to APIs without a native seed, so don't treat it as a reproduction tool.

Install

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

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

No model downloads - it's all API. You need a Zhenzhen key; the README points the lowprice nodes at the domestic store, which runs on the 7.5–8折 "everyday price" of the overseas site and explicitly carries only China-accessible models.

The gotchas that actually bite

• task is not an image. Save the Collect node's outputs, not this one's. Same trap as every node in this family.
• n is a multiplier on cost. It's the pack's biggest hidden price lever. Batch count × n = total billed calls. Plan the math before you queue 10 nodes with n=10.
• Resolution is fixed by the model. There's no resolution dropdown here - GK V1.5 decides. If you need 2k/4k, that's the GK V2 node's job.
• Metered cloud, metered wallet. The "low price" is real (that's the whole pitch of the pack), but every slot is a paid API call and your prompts plus reference images leave the machine. This is a normal, honest trade for a closed model you can't run locally - just don't confuse it with local generation.

The whole concurrent family is a convenience layer over the same API. It doesn't make the model faster per call; it makes you faster because you stop waiting one-at-a-time.