并发提交｜Seedance 文生视频 (Text to Video)

This is the concurrent-submit version of the pack's plainest video node - Seedance 2.0 text-to-video. Six models (standard/fast/mini × domestic/global), prompt in, video out, no image or audio slots to fuss with. If you're new to the pack's concurrency system, this is the node to learn it on, because it's the one with nothing extra to trip over.

How the system works, once, so it stops being confusing: a "并发提交" (concurrent submit) node doesn't generate in place like a normal Comfy node. It validates your parameters, then pushes the job onto a background video thread pool and hands you a SEEDANCE_VIDEO_FUTURE - a handle that says "a video is being made over there." You connect that handle to the SeedanceConcurrent_Video_Await receiver, which waits on up to ten handles in parallel and returns the finished VIDEO values in the order you connected them. So to generate five prompt variants of the same scene at once: five of these nodes, one receiver.

The inputs that matter

• model - six choices: seedance-2.0-{standard,fast,mini}-t2v and the three -global- versions. global- runs on the overseas channel; standard/fast/mini are quality tiers. Fast and mini are the ones to reach for when you're testing prompts before spending on standard.
• prompt - up to 20480 chars. Required, obviously - there's nothing else to drive a t2v.
• seconds - 4 to 15, or -1 to let the model pick the duration.
• resolution - 480p through 4k, plus native1080p / native4k. The native tiers are Standard-only - pick fast/mini and validation will reject them (they're not upscaled output tiers, they're native-resolution renders).
• ratio - the seven usual options, adaptive default.
• generate_audio - whether to generate voice-over/sound effects.

Plus api_config (or SEEDANCE_API_KEY / config/.env), skip_error (placeholder video on failure instead of killing the batch), and seed - -1 is random, and fixed makes ComfyUI reuse cached output when nothing changed, which is your money-saver on re-runs.

How it runs

Standard wrapper machinery: submit, poll, download, MP4-header check. Failures isolate per slot - raise mode on the receiver stops and names the slot; placeholder keeps the rest of the batch alive. Since there are no reference assets to upload, this is the fastest node in the pack to start.

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 model weights to download - the inference runs on the gateway's API. Don't skip the API key step.

Where people trip

• native1080p/native4k on fast/mini - rejected. Those are Standard-tier only.
• Long prompts on mini - mini is the cheap tier; don't expect flagship quality. Use it for coverage, standard for the keepers.
• Placeholders in the output - skip_error is doing its job; a placeholder means that specific slot failed, and the receiver's status_json tells you which and why (with secrets scrubbed).

Standard API-wrapper fine print: every clip is a metered cloud call, your prompt leaves the machine, and result links are temporary. Save what you keep.