Concurrent Submit | Zhenzhen Doubao Seedream4.5

Seedream 4.5 is the fast, character-consistent member of ByteDance's image family - the README notes it beats the Nano Banana on speed and character consistency, at 2K or 4K. That's a strong pitch, and it's wasted if you're waiting on each generation serially. This submit node is the concurrent version of zhenzhen-doubao-seedream-4.5: same doubao-seedream-4-5-251128 call, same resolution and aspect-ratio controls, but each job goes into the shared 30-worker image pool and returns a task handle immediately.

It's the natural choice for batch concept art or campaign variation work: you want the same character prompt at 1:1, 16:9 and 9:16, at 2K, and you'd rather they all render at once. With a 30-deep image pool and a fast model behind it, a whole grid lands in the time of one blocking call.

How it works

Auto-generated concurrent wrapper, same as every submit node here: the class copies the base node's inputs at import time, wraps its generate_image() 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: prompt, model (default doubao-seedream-4-5-251128), response_format (b64_json default - switch to url if you're generating many and want lighter payloads), and resolution (2K or 4K). Optional: aspect_ratio (ten options incl. Custom with width/height up to 8192), apikey, image1-5 (reference images for editing), sequential_image_generation, max_images (up to 15), seed (default -1), watermark (off by default here, unlike some sibling nodes), stream, 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 - Seedream runs on ByteDance 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. Seedream-specific: resolution is the cost lever, and 30 concurrent 4K images is a bill you'll notice - run a 2K batch first to validate prompts, then scale to 4K for the keepers. And if you're editing with image1-5 references, max_images multiplies output and cost; keep it at 1 until you know the references behave. The response_format choice matters at batch scale too - b64_json blobs for 30 4K images are heavy; url is lighter and the URLs are in the output anyway.