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

The image-edit workhorse of the pack, turned into a flood. zhenzhen-image-gk-v2-edit-lowprice is the GK V2 model's editing mode - take reference images, change them. This Concurrent Submit wrapper runs it inside the shared 30-slot image pool, so a shelf of reference images can all get their edit in flight at once. If you've ever wanted to run the same "turn this into a product shot" instruction across twenty photos, this is the node.

GK V2 is a step up from V1.5 in a few concrete ways: it takes three reference images per call instead of one, it offers 2k (not just 1k), and it has a native nsfw_check toggle - which tells you something about which crowd this pack serves. The editor's README leans on these GK nodes for e-commerce and "出海" (overseas-selling) product shots, where consistent batch editing is the whole job.

How the wrapper works

Same mechanism as every ComflyConcurrent_..._Submit node: it's generated by the pack's ComflyConcurrent.py, it mirrors the original node's inputs, validates them, and hands the job to a bounded thread pool - 30 workers for images, 10 for video. The output is a task (COMFLY_IMAGE_FUTURE), not an image. That task feeds into Concurrent Collect Images (30), which waits for everything and returns real IMAGE tensors in connection order, plus a status string.

The clever bit is the pool is shared: your GK V2 edits, your G-2 text-to-image runs, and your Z-Image jobs all queue into the same 30 image workers. Mix and match freely; the collector doesn't care which node produced which task.

Inputs that matter

• prompt - the edit instruction. This is the whole job: "change the background", "make it glossy", whatever. Multiline.
• aspect_ratio - the widest choice list in the family: auto, 16:9, 19.5:9, 20:9, 1:1, 2:3, 3:2, 3:4, 4:3, 9:16 and friends. The phone-ratio options (19.5:9, 20:9) are there because this model gets used for short-form vertical video thumbnails.
• resolution - 1k or 2k. Higher costs more; for batch product edits, 1k is usually plenty.
• n - images per call, 1–10. Multiplies the bill.
• nsfw_check - a boolean the model passes to the API. Default off. If you're doing anything the API would refuse, the refusal comes from the model's own moderation, not this flag - this toggle just asks for the check.
• image1–image3 - up to three reference images. Leave empty and the edit model has nothing to work on; it'll error.
• api_config - from zhenzhen-seedance2-low-price-api-settings. Domestic store key (api.seedance.nz).
• seed - cache-control only; not sent to APIs without a native seed.

Install

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

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

Nothing to download model-wise - it's an API client end to end. You need a Zhenzhen key; the README is explicit that these lowprice nodes are the China-facing store at 7.5–8折.

Where it trips people

• task → Collect, never Save. The collector's IMAGE outputs are what you save. Connecting the future to anything else just gives you a mysterious type error.
• Empty edit is an error. Wire at least one reference image in before you queue a batch - a dozen queued edits with no image means a dozen failed slots.
• n is a cost lever. Batch size × n = billed calls. A 10-node batch at n=10 is 100 API charges even though it felt like one click.
• Cloud moderation applies. nsfw_check or not, the underlying API enforces its own filter and you can't patch around it from the node. If the edit keeps getting refused, the model refused it - not the node, not your settings.

Reach for this when you have volume and a consistent instruction. One-off single edits are what the plain zhenzhen-image-gk-v2-edit-lowprice node is for; this is the version that turns your batch into a queue.