Vidu Q3 Turbo

Vidu Q3 Turbo is the current-generation Vidu on Runware, and the first thing you'll notice is that this node is opinionated: width and height are required inputs. Nobody else in the Vidu lineup demands that - the others default you to 1024×1024 and let you move on. Q3 Turbo makes you declare your canvas up front, which is actually a subtle signal about what generation three is: a model that treats resolution as part of the brief, not an afterthought. Set it, commit to it, generate.

The other headline is providerSettings.vidu.audio, a single checkbox that generates audio with the video. That's the big capability jump in this generation. Local video tools treat audio as a separate, painful pipeline - you generate motion, then bolt sound on with a different model. Q3 Turbo can do it in one shot, native audio straight out of the clip. Whether you use it depends on your use case, but having the checkbox is a genuinely new option in this pack's Vidu line.

How it works

Pack-standard machinery under the hood: the node (vidu:4@2, videoInference) builds a request, uploads any frameImages you wire in via the SDK, sends it over REST, and returns a native VIDEO. positivePrompt is optional here (unlike Q2 Turbo), so you can lean on starting frames if you want. duration is a plain integer, default 5 seconds, with frames = duration × the model's fps. Notably, the movement-amplitude and background-music knobs from the Q1/Q2 nodes are gone - the Q3 control surface is tighter, with the audio checkbox as the one provider-specific option.

The inputs that matter

• width / height (required) - declare your output size. Defaults are 1024×1024, but the node makes you own the choice.
• frameImages - starting frames for image-to-video.
• positivePrompt - optional; describe subject and motion.
• providerSettings.vidu.audio - the toggle that makes this generation interesting: native audio generation in the same pass.
• duration - clip length, integer seconds, default 5.
• seed / numberResults - reproducibility and variations.

Standard pack tail follows: outputFormat (MP4), outputQuality, safety toggles, ttl. One video output.

Install and API key

Install once for the whole pack:

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

Restart ComfyUI (or search Runware in ComfyUI Manager). You'll need an API key from runware.ai/api-keys, set in ComfyUI Settings → Runware API key, exported as RUNWARE_API_KEY, or stored via runware auth login.

Gotchas

The required dimensions are the easy trap: this model genuinely cares about the ratio you give it, so pick the ratio for your target (vertical for Reels/Shorts, landscape for YouTube) before you queue rather than cropping afterward. On the audio checkbox: it's one button, but it roughly doubles the compute of the run, so expect the bill to reflect it - and the generated audio is what the model thinks fits, not a mix you control, so if you need a specific voiceover, generate silent and add your own track later. Also worth noting: as the newest model in the line, Q3 Turbo is where Runware's catalog is thinnest on documented behavior - if a setting misbehaves, it's often the model's youth talking, not your wiring. Start conservative, one clip at a time.