并发提交｜Midjourney 视频

Midjourney's image-to-video is the part of MJ that people actually lean on these days - you generate a still you love, then ask for the motion. This node is the concurrent-submit wrapper around that, and it's deliberately split from the image one: the pack keeps Midjourney image futures and video futures on separate submit nodes so the receiver types don't get confused. You generate images with the image submit node; you animate them with this one.

The wrapper mechanics are pack-standard: parameters are validated, then the job runs on a background video thread pool and returns a SEEDANCE_VIDEO_FUTURE. Wire it into SeedanceConcurrent_Video_Await, which waits on up to ten video futures and hands back finished VIDEO values in slot order. Animate three stills from the same shoot as a batch - three of these, one receiver.

The inputs that matter

• operation - basically midjourney-video (image-to-video / start-end). The node's dropdown is the video action; everything else is the parameter set that action needs.
• image1 (local) or image_url1 - your source still. image1 … image4 exist as slots, but i2v really wants the first one.
• end_image / end_url - the optional end frame. Set one and the request flips into start-end video mode automatically. That's the cleanest way to get a loop that actually closes.
• video_type - vid_1.1_i2v_480, vid_1.1_i2v_720, or the two _start_end_ variants. This is the resolution/frame-mode switch.
• motion (low/high), animate_mode (manual/auto), batch_size (1/2/4) - the motion tuning knobs.
• prompt - optional, but you'll get better motion than pure "animate this" for most clips.
• The standard MJ chain fields - task_id, index (0–3 for task-reuse video), custom_id - plus speed, size, version, seed.

Plus api_config, skip_error, and the cache seed.

How it runs

The wrapper uses the same Midjourney client as the image node but targets the video action, polls to terminal state, and downloads the clip (with the MP4-header check the pack applies to all video results). Because the video result family has its own submit node, the future it returns is typed for the 10-slot video receiver - you can mix it with Kling, Seedance, Hailuo futures in the same receiver if you're feeling ambitious.

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

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

Where people trip

• Wrong node, wrong receiver - grab the image submit node when you meant video and you'll hand an IMAGE future to a video receiver and get a type error. They're separate for a reason.
• index on task-reuse video is 0–3, not 1–4 - a different convention from image actions; the tooltip calls it out.
• Start-end needs both frames - if you set an end frame, expect validation to require the pair, and pick a _start_end_ video_type to match.

Standard pack caveats: metered per clip, prompt and stills leave the machine, and links are temporary. Save the clips you keep.