并发提交｜Vidu Q3 视频生成

Vidu Q3 is the pack's video model from ShengShu - and it's the one where you should pay attention to the README before you commit a workflow to it. This node is the concurrent-submit wrapper over the whole Q3 spread: thirteen models covering text-to-video, image-to-video, start-end frame animation, and reference-to-video, with pro/turbo/fast tiers.

Concurrency is the pack standard: validate parameters, push the job to a background video thread pool, return a SEEDANCE_VIDEO_FUTURE. Wire that into SeedanceConcurrent_Video_Await, which waits on up to ten video futures and returns finished VIDEO values in slot order.

The inputs that matter

• model - 13 choices. t2v takes prompt only; i2v takes image1 as the first frame; start-end takes image1 + image2; r2v takes however many of image1…image9 you connect, compacted to order.
• prompt - required for t2v, per-model for the rest, up to 20480 chars.
• seconds - 4 to 15, submitted as a string.
• ratio - forwarded as metadata.ratio unless adaptive.
• resolution - default (leaves the API default alone), 720p, or 1080p.
• seed - -1 = random; a non-negative value gets forwarded as metadata.seed - this is one of the models in the pack that genuinely accepts a seed.

Plus api_config and skip_error (placeholder video on failure). The cache seed applies too.

The honest warning about r2v

Same story as Kling's r2v: at last testing, the Vidu reference-to-video models (vidu-q3-r2v, vidu-q3-mix-r2v, vidu-q3-ad-r2v, vidu-q3-drama-r2v) were returning an upstream 502 invalid_upstream_response at submit time. The t2v, i2v, and start-end paths were verified working. That's an upstream channel issue, not something you can fix on the node - if you need reference-based Vidu video, wait for the channel or use i2v with a composed first frame.

How it runs

Standard wrapper machinery: reference images upload automatically, submit → poll → download, MP4-header check before a result counts. One slot failing isolates: raise mode on the receiver stops and names it; placeholder keeps the batch going.

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

• r2v 502s - expected at last check. Build around it or it'll look like your workflow is broken when it isn't.
• seed is real here - unlike the cache-only seeds on some siblings, a non-negative Vidu seed is forwarded to the API, so you can actually reproduce a look.
• resolution=default means "don't send it" - if you want a guaranteed resolution, pick 720p or 1080p explicitly.

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