Vidu Q3

Vidu is the video model from Shengshu AI, the Beijing studio that's been a genuine contender in the hosted text-to-video race - the company that also had a hand in some of the strongest open-weights video research. Q3 is their current generation, and this node exposes it through Runware's cloud the same way every other model in the pack is exposed: one node, a prompt, and a VIDEO out. If you want to A/B Vidu against PixVerse or LTX from the same workflow, this is how you do it without juggling a half-dozen sign-ups.

The reason you'd specifically reach for Vidu rather than a more famous name: it's consistently strong on subject consistency and motion coherence - the things that make generated video look less like a glitchy dream sequence and more like footage. It's not the fastest or the cheapest tier, but it's the one a lot of people land on after trying the flashier options, because the clips hold together.

Inputs that matter

positivePrompt is required. The practical set:

• frameImages - first-frame image-to-video, the standard starting point for character-driven clips. Wire an IMAGE in and Vidu animates from it.
• duration - 1 to 16 seconds, default 5. The 16-second ceiling is longer than most hosted video models offer, which is genuinely useful for scenes that need room to breathe - and a direct line on your per-run bill.
• size - the resolution/aspect dropdown from 360p to 1080p, including "from input" variants. 1080p is where the cost jumps; draft at 720p.
• providerSettings.vidu.audio - generate audio along with the video. Off by default; on when you want a finished clip in one pass.
• seed - note the quirk: default is 1 (not 0), and the range starts at 1. Don't let the odd default confuse you; any value reproduces the run.

negativePrompt, numberResults (up to 4), outputFormat (MP4/WEBM/MOV), outputQuality, and safety are the standard tail.

How it works

The node sends vidu:4@1 as a videoInference task through the Runware SDK. The finished clip returns as a native VIDEO socket - Runware saves it and hands back the VIDEO type, which feeds SaveVideo or the other native video consumers. Cloud-only, pay-per-run, cost on the title bar after each queue.

Install and gotchas

cd ComfyUI/custom_nodes
git clone https://github.com/Runware/ComfyUI-Runware
pip install -r ComfyUI-Runware/requirements.txt

Restart; API key via Settings → "Runware API key", RUNWARE_API_KEY, or runware auth login. Paid, minimum top-up.

The honest warnings: hosted video is not cheap, and 16-second 1080p clips with audio are the fastest way to burn through a top-up - build your shots short and scale up only the keepers. There's no offline mode and no weights to download, so "the API is slow tonight" is something you'll just live with. And don't expect Vidu Q3 to be the best at everything: it's the consistency pick, not the special-effects pick. If you want camera-template gimmicks, PixVerse has those; if you want coherent, well-behaved motion, this is your node.