并发提交｜Seedream / Dola Seedream 图像生成/编辑

Seedream is ByteDance's image model, and it's one of the "no local door" models the KB's external-api-nodes doc uses as its canonical example of why API nodes exist at all. Through this pack you get two flavors: the domestic seedream-v5-pro and the overseas dola-seedream-5.0-pro. This node is the concurrent-submit wrapper around both, and because the image receiver handles thirty futures, it's the node for turning a batch of prompts into a wall of candidates.

The wrapper pattern is pack-standard: validate parameters synchronously, push the job to the background image thread pool, return a SEEDANCE_IMAGE_FUTURE. Feed it into SeedanceConcurrent_Image_Await (up to thirty image futures), and the receiver returns finished IMAGE values in slot order. The README even shows mixing Seedream, Image G, and Nano Banana futures into the same receiver - this node plays fine in a mixed batch.

The inputs that matter

• model_family - seedream-v5-pro (domestic) or dola-seedream-5.0-pro (overseas). The family decides the actual model, and here's the neat part: no reference images attached = text-to-image; reference images attached = image editing. The node switches request shape automatically.
• prompt - 5 to 2000 characters, required.
• resolution - 1k / 2k (API presets) or custom.
• width / height - only used in custom mode; 240 to 8192, stepped by 8.
• output_format - png or jpeg.
• image1…image10 - the optional edit reference bank. Connect any and the job becomes an i2i edit.

Plus api_config, skip_error (placeholder image on failure instead of killing the batch), and seed as a ComfyUI cache seed - fixed reuses cached results so an unchanged re-run doesn't re-bill.

How it runs

Standard wrapper machinery: the domestic path routes to seedream-v5-pro-t2i/i2i, the overseas to dola-seedream-5.0-pro-t2i/i2i. Reference images are uploaded automatically, the worker does submit → poll → download, and you get a plain IMAGE tensor back. The image pool defaults to 30 workers (SEEDANCE_IMAGE_CONCURRENCY to tune).

Installing

# ComfyUI Manager: search "ComfyUI Seedance" → install → restart.
# or:
comfy node install seedance
# or:
cd ComfyUI/custom_nodes
git clone https://github.com/T8mars/ComfyUI_Seedance.git
cd ../.. && python -m pip install -r custom_nodes/ComfyUI_Seedance/requirements.txt

Only dependency is requests; no weights. API key via Seedance API Config, SEEDANCE_API_KEY, or config/.env.

Where people trip

• Editing vs generating is implicit - there's no "mode" switch. Connect a reference image and you're editing; don't and you're generating. Forgetting an image that's still wired into a slot is the classic "why is it editing my photo" moment.
• Custom resolution fields do nothing in preset mode - width/height only fire when resolution is custom. The tooltips say so; trust them.
• Short prompts - under 5 characters gets rejected at validation. Seedream wants a real sentence.

Pack-wide fine print: prompts and reference images leave the machine, every image is metered, and links expire - save what you keep.