Vidu Img2Video

Vidu is Shengshu AI's video family - the q3 series is their current fast-quality line - and UCloud hosts it on the ModelVerse API. Vidu Img2Video takes a starting image and turns it into a video clip, with one thing the other video nodes here don't offer as freely: up to 16 seconds of output, plus a bgm toggle to get background music baked into the clip. If you want longer, livelier image-to-video without the Veo price tag, this is the workhorse.

The shape of a good workflow: generate a strong still (the pack's Qwen Image or NanoBanana nodes are right there), feed it in, and let Vidu add motion. Because it's hosted, you also don't need the GPU that a local i2v model would demand.

The inputs that matter

• client - standard wiring from Modelverse Client or Secret Client.
• model - five options: viduq3-pro (default) and viduq3-turbo (both 1–16s), plus the older viduq2-pro, viduq2-turbo, and viduq2-pro-fast (1–10s). The tooltips are honest: q2-pro is "good results," q2-turbo is "fast generation." Start with q3.
• duration - 1 to 16 seconds, but only if the model is q3-series. The node validates this: pick 12 seconds with a viduq2 model and it refuses with a clear error.
• resolution - 540p, 720p (default), or 1080p.
• movement_amplitude - auto, small, medium, or large. This is your motion dial: small for subtle, large for dramatic. auto is a sane default.
• first_frame_image / first_frame_url - one required, never both (the node raises an error otherwise). Wire your still into first_frame_image.
• prompt - optional, max 2000 characters. Describe the motion, not the whole scene.
• seed and bgm - seed for reproducibility; bgm (default off) adds background music to the output.

Outputs are url and task_id; url goes to Modelverse Preview Video to save and preview.

How it works

The node base64-encodes your first frame, submits a task flagged vidu_type: img2video, and polls every 5 seconds until the clip renders. Same async task pattern as the other video nodes here. It's metered and it can take a couple of minutes for the long clips, so budget accordingly - and remember duration is a bill multiplier, not just a quality setting.

Install

The pack installs once for all its nodes: ComfyUI Manager → search "ComfyUl-UCloud" → Install → restart, or git clone https://github.com/ucloud/comfyui-plugin into custom_nodes. Nothing heavy to download - the "model" lives on UCloud's servers, and your only setup is an API key from the Modelverse console.

One thing I'd actually reach for this over Wan for: when I want more than 5 seconds. The Wan nodes are stuck at 5; Vidu q3 stretches to 16. For a draft-and-iterate loop, the turbo models keep cost down; for the keeper, viduq3-pro at 1080p is the one.