并发提交｜Vidu Q3 短剧成片

This is the pack's most unusual node, full stop. Instead of a prompt, you give Vidu Q3 a short-play script, a script name, and a set of character/scene/prop asset images - and it renders a finished short-form clip (8–12 seconds, 1080p) from that material. It's aimed squarely at the vertical-drama crowd, and the 9:16 default aspect ratio tells you everything about who this is for. This is its concurrent-submit version.

Mechanics are the pack standard: validate, push the job to the background video thread pool, return a SEEDANCE_VIDEO_FUTURE. Feed it into SeedanceConcurrent_Video_Await and it returns the finished VIDEO in slot order alongside whatever else you're running.

The inputs that matter

• model - vidu-q3-drama-short-play or vidu-q3-ad-short-play (drama vs ad spot).
• prompt - the script content itself, forwarded as the script prompt. This is where the scene-by-scene story lives.
• script_name - forwarded as metadata.script_name; the upstream scriptName.
• resolution - locked to 1080p; there's one choice and that's it.
• duration - 8 to 12 seconds.
• aspect_ratio - 9:16 or 16:9. The vertical default says it all.
• style - up to 30 characters ("realistic" default).
• asset_type - applies one type (character, scene, or prop) to all connected assets. There's no per-asset typing, so keep each batch single-type.
• asset_name_prefix - names are auto-built as "<prefix> 1", "<prefix> 2", etc.
• asset_description - one description applied to every connected asset.
• asset_image1…asset_image14 - the reference bank; at least asset_image1 is required. These get uploaded and written into metadata.assets[].image_uri.

Plus api_config, skip_error, and the cache seed.

The honest warning

The README is explicit: the short-play models were still returning an upstream 502 at submit time when last tested. The node is built and validated against the documented request shape, but the actual channel was broken - treat this one as "waiting on upstream," not as a finished pipeline you can rely on today. Check the pack's changelog/README before building a whole drama workflow around it.

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

• Single asset_type for everything - mix a character and a prop in one batch and both get tagged the same. Split them into separate jobs.
• No assets - at least one asset image is required; validation bounces the job without it.
• The 502 - if it hits you, it's upstream, not your workflow. Watch the pack's release notes for the channel coming back.

And the standing pack reminder: this is a metered cloud call, your script and reference assets leave the machine, and the links are temporary. Save the renders you keep.