Vidu Q3 短剧成片

Most video nodes ask you to write a prompt. Vidu_Q3_ShortPlay asks you to write a script. It's the "短剧成片" node - short-drama finishing - and it takes a script, a cast of character/scene/prop reference images, and turns them into a finished short-form clip, which is a whole different job from a one-shot text-to-video generation. You hand it a story and assets; it hands back a produced take.

Vidu is a closed Chinese video model, so there's no local option at any VRAM - an API wrapper is the only door, and this one routes through the pack's reseller endpoint (api.seedance.nz). That framing matters twice here: you're paying per clip through a third party, and you're also at the mercy of the upstream's availability, which for this specific node has been shaky (more below).

How it works

Pick one of the two models - vidu-q3-drama-short-play or vidu-q3-ad-short-play (drama vs. ad) - and the node assembles the request. Your prompt is the script content, forwarded as the script; script_name goes into metadata.script_name. The reference images you connect (asset_image1 through asset_image14) get uploaded and labeled as assets, with the asset_type (character, scene, or prop), asset_name_prefix, and asset_description fields deciding how each one is described to the model. That's the mechanism worth understanding: this isn't img2img with a single first frame, it's "here are my characters and locations, now act out this script." At least asset_image1 is required - the node is strict about that.

Resolution is locked to 1080p (the API requires it), duration is 8–12 seconds, aspect is 9:16 or 16:9 (you want 9:16 for short-drama, that's the format), and style caps at 30 characters.

Inputs and outputs that matter

• prompt - the script. This is the creative input.
• script_name - a name for the piece, passed through as metadata.script_name.
• model - drama or ad short-play.
• duration, aspect_ratio, style, resolution (fixed 1080p).
• asset_image1–asset_image14 - reference character/scene/prop images, labeled via asset_type, asset_name_prefix, asset_description.

Outputs are the standard video quartet: video (downloaded and decoded to ComfyUI VIDEO), video_url, task_id, and response.

Install

This is one node in the T8mars/ComfyUI_Seedance pack, so install the pack once:

# ComfyUI Manager: search "ComfyUI Seedance", or:
comfy node install seedance

Or git clone https://github.com/T8mars/ComfyUI_Seedance.git into ComfyUI/custom_nodes and pip install -r requirements.txt (just requests). You'll need an API key from api.seedance.nz/console, provided via the Seedance API Config node, the SEEDANCE_API_KEY env var, or a config/.env file.

The honest caveat

The pack's own changelog (v0.2.3) says both short-play models were returning an upstream 502 invalid_upstream_response at the submission stage when the node was tested - a Vidu-side problem, not a config error on your part. The example workflow exists and the node is fully wired, but this is the one node in the Vidu family where "it might just not go through" is a live possibility depending on the upstream's current state. If it 502s on you, check the pack's release notes to see whether it's been cleared before you spend time debugging your own setup. When it does work, budget accordingly: an 8–12 second 1080p drama clip with assets is one of the pricier things this pack can do.