Vidu2 Text-to-Video Generation

The first-generation Vidu nodes lock you at 5 seconds and 1080p and call it a day. Vidu 2 is where the training wheels come off: the duration slider actually works (1 to 10 seconds), you get a real choice of resolutions, five aspect ratios, and even an optional background-music toggle. It's the "I've tried the free taste, now I want the real thing" node.

The model is viduq2 - note there's just the one choice here, whereas the Vidu 2 image-to-video node gives you a pro/turbo split. Text-to-video on Vidu 2 is one model, one job: prompt in, video out, no start frame required.

What you set

• prompt - required, max 2000 characters. Prose works; Vidu rewards describing motion and camera behavior explicitly.
• duration - 1 to 10 seconds. This is the headliner. The longer you go, the more the model has to invent, so 8–10 second clips are where quality and cost both climb. The price badge updates live, which makes the tradeoff visible before you queue.
• aspect_ratio - 16:9, 9:16, 3:4, 4:3, or 1:1. More choice than Vidu 1, and the portrait/social options are genuinely useful.
• resolution - 720p or 1080p. 720p is the iterate-cheap choice; 1080p is the deliverable. If you're experimenting, run 720p, lock the seed on a good take, then re-run at 1080p.
• background_music - adds generated background music to the clip. It's a boolean, it's zero-effort, and it does what it says. Turn it on when you're prototyping a look; keep it off if you're going to add your own soundtrack in an editor, because separating it out later is a chore.

The seed input defaults to 1 here (instead of 0-for-random like the Vidu 1 nodes), so the first run is already reproducible. Set it to 0 for true randomness, otherwise it's a fixed reroll button.

How it runs

Same API-node plumbing as the rest of the family: ComfyUI POSTs to Vidu's text-to-video endpoint through the Comfy proxy, polls the task, downloads the result. Needs a Comfy account with credits and an internet connection - no local GPU, no model files. Output is a VIDEO object; save it downstream or it's lost when the queue clears.

Notes worth having

Vidu 2 pricing is per-second and per-resolution, which is exactly why the 720p-then-1080p pattern is the community-standard workflow - it's the difference between a coin flip and a deliberate spend. And one thing that applies to the whole Vidu family: the content rules are the same through ComfyUI as they are on Vidu's own site. The point of using these nodes isn't looser moderation, it's that the output lands inside your graph - ready to upscale, edit, or hand to the next stage. If you came here expecting Vidu to do something it won't on its own site, you'll be disappointed.