Zhenzhen Vidu Image2Video

Vidu is ByteDance's video model family - the kind of thing that normally lives behind a Chinese consumer app and a login you can't get. Comfly_vidu_img2video is the ComfyUI relay that hands you the API anyway: an image in, a short video out, with the whole menu of Vidu options - duration, resolution, motion amount, BGM, and a voiceover system with three hundred voices - exposed as plain node inputs.

It's part of Comfyui_Comfly, so the ground rules hold: paid relay to ai.comfly.chat, one shared API key, zero local compute. The node base64-encodes your image, POSTs it to the relay's /vidu/v2/img2video endpoint, then polls the task endpoint every 10 seconds until the clip is ready. Expect minutes, not seconds.

Inputs that matter

• image - the starting frame. This drives everything; Vidu animates what it sees.
• model - six choices: viduq2-pro, viduq2-turbo, viduq1, viduq1-classic, vidu2.0, vidu1.5. Q2-pro is the current default and the one to reach for first; turbo trades quality for speed.
• prompt - optional but strongly recommended motion description ("the camera pans right as she walks toward the window").
• duration - 1–10 seconds (default 5).
• resolution - 540p/720p/1080p. Higher costs more, and the default 720p is a fine starting point.
• movement_amplitude - auto/small/medium/large. Small keeps things subtle; large is for dramatic camera moves.

Then the extras that make Vidu distinctive:

• audio - turn this on to attach a voiceover, then pick voice_language (17 languages, Chinese default) and voice_id (a 300+ entry list of named personas - male-qn-jingying, female-shaonv, and friends).
• bgm - background music on top.
• watermark - add a visible watermark; wm_position picks the corner (1–4).
• off_peak - run during off-peak hours for cheaper/faster processing.
• is_rec - a Vidu-specific mode toggle; leave off unless you know you need it.
• seed - reproducibility.

Outputs

video - a VIDEO object that wires straight into ComfyUI's Save Video node. video_url is the hosted link, task_id identifies the job on the relay (where you can also track it), and response is the JSON log, errors included.

Install

cd ComfyUI/custom_nodes
git clone https://github.com/ainewsto/Comfyui_Comfly

or ComfyUI Manager → search "Comfyui_Comfly" → install → restart. API key from ai.comfly.chat into the api_key field.

The honest take

This is the most config-heavy node in the pack, and the menu is genuinely the feature - where else do you get 300 named voices in a ComfyUI dropdown? Just don't touch everything at once. Voice + BGM + watermark + 1080p is a recipe for a slow, pricey job; start with image + prompt + default duration, then add layers. The polling loop is patient (up to ~20 minutes of attempts), so a long wait isn't a hang - it's Vidu. And remember the model family is API-only by ByteDance's design, so this relay is the only way most of you will ever touch it.