Concurrent Submit | Zhenzhen Doubao Seededit3.0

Seededit is ByteDance's image-edit model - the 3.0 version here takes an input image plus a prompt and re-renders it to match, the classic "keep the subject, change everything around it" tool that e-commerce teams use to restyle the same product shot a hundred ways. The blocking zhenzhen-doubao-seededit-3.0 node does one edit per wait. This submit node is the concurrent version: same doubao-seededit-3-0-i2i call, same image-plus-prompt contract, but each job goes into the shared 30-worker image pool and returns a task handle instead of a finished image.

This is one of the better concurrency fits in the pack, because product restyling is inherently a batch job. Same source image, five different prompts ("on a white studio background", "in a forest at sunset", "on a neon city street") - the serial version makes you sit through each one. The concurrent version fires all five and the collector hands them back in order.

How it works

Standard auto-generated concurrent wrapper: the submit class copies the base node's inputs at import time, wraps its edit call in a bounded ThreadPoolExecutor (30 image workers by default, set via COMFLY_IMAGE_CONCURRENCY), and returns a COMFLY_IMAGE_FUTURE. Collect with ComflyConcurrent_Image_Await ("Concurrent Collect Images (30)"), which waits for the batch and returns IMAGE tensors in slot order.

Required inputs: image (the source IMAGE), prompt, model (default doubao-seededit-3-0-i2i-250628), response_format (url or b64_json), size (default adaptive), and guidance_scale (5.5). Optional: apikey, seed (default -1 = random), watermark (on by default - flip it if you're getting watermarked output), and skip_error.

Installing it

One clone, whole pack:

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

Restart ComfyUI, or install "Comfyui-zhenzhen" via ComfyUI Manager. No model downloads - Seededit runs on ByteDance's hardware behind the proxy.

Common issues

Pack-level: account + credit required, 443 errors need VPN with TUN mode, first-run 500s are usually upstream - re-run. Seededit-specific: the watermark toggle is on by default, which catches a lot of people on their first batch - if your edits come back stamped, that's the cause. And guidance_scale is a batch-tunable knob: tune it once on a single blocking edit before you fire 30 concurrent ones, because a too-high value across a whole batch is an expensive lesson. Meter note applies: 30 concurrent edits are 30 paid calls.