并发提交｜Midjourney 图片

Midjourney inside ComfyUI is a weird flex - the KB's closed-source doc points out that Flux basically closed the quality gap, and the community moved on. But a lot of people still have Midjourney subscriptions and a library of MJ work, and this node is how that gets into a workflow. It's the concurrent-submit version of the pack's big 16-in-1 Midjourney node, covering the image-result operations - imagine, blend, edits, upscale, variation, reroll, zoom, pan, and the modal region repaint.

The wrapper pattern is unchanged: fill the parameters, the node validates, then the job goes to the background image thread pool and you get a SEEDANCE_IMAGE_FUTURE back. That goes into SeedanceConcurrent_Image_Await, which waits on up to thirty image futures and returns them in slot order. Thirty MJ images in one batch, all through one receiver - that's the whole point of this node.

The inputs that matter

This node has a genuinely huge field set, because the underlying 16-in-1 node does, and which fields matter depends entirely on operation:

• operation - the action dropdown; the Chinese suffix after each ID tells you what it does. midjourney-imagine for text/ref generation, blend for 2–4 image fusion, upscale/variation/reroll for the second-pass actions, modal for region repaint.
• prompt - the prompt or edit instruction; external STRING nodes feed in fine.
• speed (relax/fast/turbo), size / custom_size, version (default 8.2), quality - the standard MJ knobs, mostly at their documented defaults.
• task_id / index / custom_id - how you chain: connect a prior Midjourney node's task_id and pick the image index (1–4) for upscale/variation.
• image1…image4 (local) or image_url1…image_url4 - reference images for blend/edits; don't fill both sides of the same slot.
• mask / modal_mode - for the repaint action, region mode takes a ComfyUI mask (white = repainted); the no-mask outpaint mode is still rejected upstream per the README.

Plus the pack's standard api_config, skip_error, and seed. The seed here is real MJ seed plumbing - -1 means don't send it, and fixed on the ComfyUI side also reuses cache.

How it runs

The wrapper calls the pack's Midjourney client with an explicit action path, polls to a documented terminal state (including the MODAL state, which is a legal "waiting for params" stop), downloads the result image(s), and returns a normal IMAGE. The image pool defaults to 30 workers, tunable via SEEDANCE_IMAGE_CONCURRENCY.

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

Where people trip

• index semantics - 1–4 for image actions, but 0–3 when a Video task reuses a task. The tooltip says it plainly; get it backwards and you'll reference the wrong candidate.
• outpaint without a mask - documented upstream limitation; the pack keeps the field but the modal flow currently needs region + a mask.
• Relax vs fast - relax is the default and can queue a long time if the account is congested. If a batch seems stalled, that's likely it.

It's the API-wrapper trade all over again: your prompts and images go to the gateway, and every action is metered. With MJ specifically, remember the account itself is usually a subscription - the pack just puts a ComfyUI face on it.