并发提交｜Seedance 多模态视频 (Multimodal Video)

The multimodal model is where Seedance 2.0 stops being "text in, video out" and starts being a reference-hungry director: you feed it up to nine images, three videos, and three audio clips, and it composites a clip that respects all of them. This node is the concurrent-submit wrapper around that. It's not the newest multimodal in the pack (2.5's Multi takes 30 images), but it's the one with the tightest, most approachable reference model - and its batch version is what you want when one composition isn't enough.

Concurrency works like every submit node here: validate, push to the background video thread pool, get a SEEDANCE_VIDEO_FUTURE back. Feed up to ten of those into SeedanceConcurrent_Video_Await and it returns finished VIDEO values in slot order. Two different reference sets, same prompt, compare the takes - two of these, one receiver.

The inputs that matter

• model - seedance-2.0-{standard,fast,mini}-multi plus -global- versions.
• prompt - required, and it's how you direct the references. Address them in the prompt with @Image 1, @Video 1, @Audio 1 - that's the API's own addressing scheme, not decoration.
• image1…image9 - up to nine reference images.
• video1…video3 - reference videos; MP4, ≤50MB each.
• audio1…audio3 - reference audio; ≤50MB each.
• seconds - 4 to 15 or -1; resolution - 480p…4k plus the Standard-only native tiers; ratio - adaptive follows your material.
• generate_audio - toggle for the output's own audio.

Plus api_config, skip_error, and the cache seed (fixed = reuses cached results on unchanged runs).

How it runs

Standard wrapper machinery: all your local reference media gets uploaded automatically, the pack compacts the reference numbering to connected order (and prints a console hint when it renumbers), then submit → poll → download with the MP4 check. The upload phase is the slow part with big reference sets - the README warns the API rate-limits per-token uploads and the client retries.

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

• @Image tags that miss - numbering is by connected slot order, and gaps get compacted. Connect image1 and image3 and your second reference is @Image 2, not @Image 3.
• Oversized references - videos and audio over 50MB get bounced. Trim before you wire them in.
• Multi with no references - the model needs at least one connected asset; a prompt-only "multi" job fails validation. Use the t2v node for that.

Per-clip metering and data-leaves-the-machine caveats, now at maximum strength - you're shipping up to 15 assets to the gateway per job. Save what you keep.