Nodes/Comfyui-zhenzhen/Concurrent Submit | zhenzhen-image-g2-lowprice
ComfyUI Node

Concurrent Submit | zhenzhen-image-g2-lowprice

Fire 30 zhenzhen-image-g2 prompts at once and go make tea

By T8mars·Created about a year ago·Updated 2 days ago· 740
Concurrent Submit | zhenzhen-image-g2-lowprice
  • api_config
  • image1
  • image2
  • image3
  • image4
  • image5
  • image6
  • image7
  • image8
  • image9
  • image10
  • task
modelzhenzhen-image-g2-t2i
prompt
resolution1k
ratioadaptive
skip_errorfalse
seed0

This node is the "hit 30 buttons at once" edition of zhenzhen-image-g2-lowprice. The plain node is a one-shot cloud API call - submit a prompt, wait for the image, get a tensor back. That's fine for a single picture, but the moment you want a wall of variations, a character sheet, or a prompt matrix, waiting one-at-a-time is painful. The Concurrent Submit wrapper runs the same node inside the pack's shared 30-slot image pool, so a whole batch of them is in flight simultaneously. Real concurrency, no local GPU involved - the model lives on the Zhenzhen API, your machine is just the remote control.

It's part of the T8mars/Comfyui-zhenzhen pack ("Comfyui-zhenzhen"), a fan-run reseller of closed cloud models - think Nano Banana, Seedream, Gemini-2-era image models at reseller prices. G-2 here is the lowprice text-to-image / image-edit model, capped at 1k resolution.

The two-node pattern

Every node whose name starts ComflyConcurrent_..._Submit works the same way: it wraps the original node, validates your inputs, and hands the work to a bounded ThreadPoolExecutor - 30 workers for image nodes, 10 for video, tunable via COMFLY_IMAGE_CONCURRENCY / COMFLY_VIDEO_CONCURRENCY env vars. The output is a task (type COMFLY_IMAGE_FUTURE), not an image. You feed that task into the pack's Concurrent Collect Images (30) node, which waits for every queued task and hands back real IMAGE tensors in the order you connected them.

So: N Submit nodes → one Collect node. Your queue ends up in a status string output that reports each slot as pending, success, or failed.

Inputs that matter

  • model - zhenzhen-image-g2-t2i (text-to-image) or zhenzhen-image-g2-i2i (edit with reference images).
  • prompt - the actual text. Multiline.
  • ratio - adaptive, 16:9, 4:3, 1:1, 3:4, 9:16, 21:9. G-2 is fixed to 1k resolution, so ratio is how you pick the shape.
  • image1–image10 - reference images for the i2i model. The tooltip-free widgets hide an important rule: leave them empty for t2i.
  • api_config - wire this from the pack's zhenzhen-seedance2-low-price-api-settings node (base_url + api_key). This is the domestic Seedance2 low-price config, so it expects the key from api.seedance.nz, the China-facing store.
  • seed - this one deserves a paragraph.

The seed is a ComfyUI cache-control seed, not a sampler seed. A fixed value reuses the cached result; randomize/increment/decrement forces a new API run. The author's tooltip says it plainly: it is not sent to APIs that don't expose a native seed. So don't expect the same seed to reproduce the same image across separate runs - it can't, the model is on someone else's server.

Install

ComfyUI Manager → Install Custom Nodes → search "Comfyui-zhenzhen" → install → restart. Or by hand:

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

No checkpoints, no safetensors, nothing to download. The pack is a pure HTTP client (requests/aiohttp - the heavy stuff like torch is already in your ComfyUI env). What you do need is an API key from the author's store. The README makes the deal explicit: it's a non-profit-ish reseller keeping a 5–10% margin for hosting, no price negotiation, and the domestic site only carries domestic (China-accessible) models.

Where people get burned

  • The task output is not an image. Plug it into the Collect node, not into Save Image. This is the #1 confusion with the whole concurrent family.
  • The China store needs the China key. The domestic config endpoint (api.seedance.nz) and the overseas one (ai.t8star.org) are different sites with different keys.
  • The overseas endpoint needs a proxy. 443 errors, can't connect - the README says open the magic (VPN) in TUN mode, and to blame your VPN/router/360 firewall before the server.
  • It's metered. Every batch slot is a billed call. Blasting 30 at once is 30 charges, even if the pool makes them feel free. The "low price" is the point - check the site's current per-image rate before you fire off a 30-grid.

One light word of caution that applies to every node in this pack: it's arbitrary Python that carries your API key and phones home by design. That's the normal shape of an API-wrapper node, but it's also exactly the shape that's been weaponized once in this ecosystem - install from the official repo via Manager, and be aware your prompts and reference images leave your machine for a third-party server.

Categoryzhenzhen/Seedance2 Low Price/Concurrent Submit

Inputs (17)

NameTypeDefaultDescription
modelCOMBOzhenzhen-image-g2-t2i2 options: zhenzhen-image-g2-t2i, zhenzhen-image-g2-i2i
promptSTRING
resolutionCOMBO1k1 options: 1k
ratioCOMBOadaptive7 options: adaptive, 16:9, 4:3, 1:1, 3:4, 9:16, +1
api_configoptZHENZHEN_SEEDANCE2_CONFIG
skip_erroroptBOOLEANfalse
image1optIMAGE
image2optIMAGE
image3optIMAGE
image4optIMAGE
image5optIMAGE
image6optIMAGE
image7optIMAGE
image8optIMAGE
image9optIMAGE
image10optIMAGE
seedoptINT00–18446744073709550000Execution seed for ComfyUI cache control. Fixed reuses the cached result; randomize/increment/decrement requests a new run. This compatibility seed is not sent to APIs that do not expose a native seed parameter.

Outputs (1)

NameTypeDescription
taskCOMFLY_IMAGE_FUTURE