Vidu Text2Video

Vidu is Shengshu's video model line, and this node gives you three generations of it - viduq3-pro, viduq3-turbo, and viduq2 - behind a single model dropdown. Like the rest of this pack it's an API wrapper into UCloud's Modelverse service, so the model runs on their GPUs and you run the invoice. It's the middle-of-the-road video option in the pack: more polished than the quick-sketch HappyHorse lane, less dramatic than Sora 2, and with a turbo tier that's genuinely useful when you're iterating.

What it is

Text-to-video, full stop. Prompt in, 540p–1080p mp4 out. The interesting part is the model menu: pro for detail-rich output, turbo for speed, and viduq2 for the older generation if you have workflows tuned to its look. That's a nice way to scale cost-versus-quality within a single node.

The inputs that matter

• client - required wire from the Modelverse Client / Secret Client node.
• model - viduq3-pro (default, "good results, rich detail"), viduq3-turbo ("generates fast"), or viduq2 (legacy).
• prompt - up to 2000 characters. Vidu handles long cinematic descriptions well; use them.
• duration - seconds, default 5.
• aspect_ratio - 16:9, 9:16, 3:4, 4:3, or 1:1.
• resolution - 540p, 720p (default), or 1080p.
• guidance_scale - 7.5 default. This is CFG in the traditional sense; lower it for more creative drift, raise it for prompt-lock.
• bgm - flips background music on. Vidu's audio is a nice-to-have if you're shipping clips without a music budget.
• seed - 0 for random, or lock it.

Outputs are the standard url + task_id pair. Pipe url into the Modelverse Preview Video node to download and preview the mp4.

Installing and keying up

Same plugin as every node here. ComfyUI Manager → search "ComfyUl-UCloud" → Install → restart, or:

cd ComfyUI/custom_nodes
git clone https://github.com/ucloud/comfyui-plugin

restart and you're in. Dependencies (requests, pillow, numpy, openai, GitPython) come along via Manager; there are no weights to download because the model lives server-side. The one setup step is a ModelVerse API key from the UCloud console, entered through the Modelverse Secret Client node (name only, key stays in secrets.json) or a config.ini MODELVERSE_API_KEY=... line.

Common issues

• Turbo vs pro look different - they're different models, not a speed slider. If turbo results feel thin, that's the tier, not a broken setting.
• bgm adds nothing - some prompts just don't inspire a track; it's optional for a reason.
• Preview shows nothing - the video node returns a URL string; it must pass through Modelverse Preview Video to become visible.
• "API key is not set" - empty client node. Check it first, always.
• Long waits - like all cloud video here, jobs poll every 5 seconds up to ~15 minutes. The pro tier on 1080p is the slow one.

Practical tip: draft in turbo, and only when a take survives your eyeball test, re-roll it in pro at final resolution. You'll halve your spend and your patience budget at the same time.