Concurrent Submit | Zhenzhen Jimeng API

Jimeng (即梦) is ByteDance's generation suite, and this node wraps its image API the way the pack wraps everything: same call as the plain ComflyJimengApi node, but handed to a shared pool of 30 workers so you can batch a whole grid of generations at once instead of blocking the graph one image at a time.

The genuinely unusual thing here is the built-in logo stamper. The node can overlay a logo - text, language, position, opacity - onto every generated image as part of the API call, which is the kind of thing most API nodes make you do in post with a separate compositing step. If you're generating product shots or branded content where every output needs a watermark, that's a workflow shortcut you didn't know you wanted.

How the wrapper works

Underneath, the plain ComflyJimengApi node posts your prompt to Jimeng's image endpoint through Zhenzhen's proxy, uploading a reference image if you connect one, and returns the rendered image. The Submit wrapper keeps every input but returns a single task of type COMFLY_IMAGE_FUTURE, deferring the call to the image pool (default 30, COMFLY_IMAGE_CONCURRENCY env var).

The task must flow into ComflyConcurrent_Image_Await, which waits for all connected tasks and returns image_1 … image_30 in submission order plus a status JSON. failure_mode (fail_fast / placeholder) controls whether one failure aborts the batch or blanks the slot.

Inputs that matter

prompt is required. Then there's the odd quartet that's really Jimeng's signature: scale (default 2.5, the guidance-like control), seed (default −1, so random until you set one), and width/height (default 1328×1328 - the 2×2 style of grid output Jimeng likes; min 512, max 2048, step 8). The size step of 8 catches people: 1327 will silently snap.

The logo block is optional: add_logo (false), then logo_text (what it says), logo_position (右下角/左下角/左上角/右上角 - bottom-right is default), logo_language (中文/英文), logo_opacity (0.3). The position/language values are literally Chinese in the dropdown; that's the pack's UI, not a bug. image/image_url enable image-reference generation (which costs extra upload tokens per the README), and use_pre_llm runs a prompt-enhancement LLM pass first. api_key is optional once API Settings is configured.

Install and gotchas

Pack install as always: ComfyUI Manager → search Comfyui-zhenzhen, or git clone https://github.com/T8mars/Comfyui-zhenzhen into ComfyUI/custom_nodes, restart. No local models.

The cost model is per-image and the README is upfront that reference-image generation adds upload token costs beyond the render. Jimeng's output grid means one "image" can be a 2×2 set, so when you think you got one image you sometimes got four tiles - check before assuming a download is corrupted. A 500 is usually upstream ByteDance load; re-run. And if a slot shows failed at the Await node but the image clearly exists, the download timed out - the site's task page has it.