How to prompt Vidu Q3: the controls that do nothing, and the words that do

Vidu Q3 is Shengshu's joint audio-video model: lip-synced dialogue, voiceover, effects and music generated in the same pass as the picture. On LUVI it runs as Q3-Pro and Q3-Turbo for text, image and start-end frames, plus reference-to-video on Q3 and Q3-Mix. Clips run from 1 to 16 seconds at 540p, 720p or 1080p. On September 22, 2026 a five-second clip at 720p quoted 510 credits on Q3-Turbo and 1,063 on Q3-Pro, with Q3-Pro at 1080p coming to 1,275. What makes this family unusual is which of its controls are real.

Which controls do nothing on Q3?

The style control, by the vendor's own admission. Vidu's parameter table carries a note on style: "This parameter is not effective when using q2 or q3 models." The movement_amplitude field that appears throughout Vidu's sample requests is not in the documented parameter list for this endpoint either, and LUVI's prompting guide for the family treats it as inert on Q3.

Neither appears in the models' parameter contracts on LUVI, which is consistent: there is nothing there to set. The practical consequence is the whole point of this guide—on Q3, look and motion energy are not settings. They are sentences.

So how do I set the look and the motion?

By naming both in the prose, as plainly as you would name the subject.

• The look goes in as a style phrase, usually up front: "anime", "ultra-realistic fashion photography style", "grainy documentary". Vidu's own example prompt opens exactly that way: "In an ultra-realistic fashion photography style featuring light blue and pale amber tones, an astronaut in a spacesuit walks through the fog."
• The motion energy goes in as a description of how much movement there is: "barely moves", "shifts her weight once", "sprints". Without it the model picks its own amount.
• The camera goes in with film terms—pans, push-ins, tracking shots, "slow aerial drone shot", "static locked-off". A named move is steadier than an implied one.

In an ultra-realistic documentary style, a beekeeper in a pale veiled suit lifts a wooden frame from a hive in late afternoon light, moving slowly and deliberately as bees drift around her gloves. One continuous shot, the camera pushes in slowly toward her hands. Bees hum steadily, smoke hisses from a canister, a distant tractor passes on the road.

Multi-shot is native on this family—Vidu calls it smart cuts—so when the story has several beats, write them in order with their own framing and their own sound and let the model place the cuts. When you want a single take, say so, as the example does.

How do I write the sound?

As content, in the same paragraph as the picture. The audio switch is on by default; Vidu's documentation describes it as the "direct audio-video generation capability", and turning it off "outputs silent video". Note that it applies to the Q3 models only.

There is a trap next to it. LUVI's Q3-Pro contract exposes a bgm control with a default of on, but Vidu's own parameter table says BGM "does not available in q3 models". Do not rely on that switch to give you a score: describe the music in the prose instead, the way you describe the effects.

Dialogue works in English, Japanese and Chinese. Another language may render poorly, so keep such a line very short or carry the beat without on-camera speech. Label each speaker uniquely when more than one talks, and give every line its own sentence.

What else should I set deliberately?

Four things, and one of them surprises people.

• aspect_ratio defaults to 4:3, not 16:9. If you are producing for a widescreen timeline or for vertical, set it every time; the values are 16:9, 9:16, 4:3, 3:4 and 1:1.
• duration runs from 1 to 16 seconds, 5 by default. Sixteen seconds is room for a short sequence, not for one long action—a single action stretched across it tends to slow down rather than develop.
• resolution is 540p, 720p or 1080p, and it moves the price: 1,063 credits at 720p against 1,275 at 1080p on Q3-Pro, for the same five seconds.
• seed exists, so a shot can be held steady while you change one clause.

The prompt itself is capped at 5,000 characters—generous, but a clip prompt is a paragraph, not a script.

How we checked

• Models: Vidu Q3-Pro Text-to-video and Q3-Turbo Text-to-video, with the image, start-end and reference models read from the catalog.
• Settings: five-second clips at 720p and 1080p.
• Date and what was read: September 22, 2026—the models' parameter contracts and live credit quotes in LUVI, LUVI's own prompting guide for the Vidu family, and Vidu's text-to-video API documentation.
• Results: 510 credits on Q3-Turbo and 1,063 on Q3-Pro at 720p; 1,275 on Q3-Pro at 1080p. Neither model exposed a style or motion-amplitude control. No outputs were generated for this post, so it makes no claim about output quality.
• Drawbacks: the bgm discrepancy is read from two documents rather than measured on a render, so treat it as a reason to write the music into the prose, not as a proven silence.

In LUVI

Open a model from the Vidu Q3 family page, and before anything else check the aspect ratio—the 4:3 default is the setting people forget. Write the style, the motion and the sound into the paragraph, then let the feather button beside the prompt box run LUVI's prompting guide for the model. If you are deciding between this family and one where the audio is a toggle, the help center's list of which models generate sound with the video is the shortest way to compare.

New accounts start with 200 credits, and verifying a phone and a card adds up to 1,500 more. You can create a LUVI account and hear what a written-in soundtrack does.

Sources

• Text-to-video API reference — Vidu, official platform documentation. Accessed September 22, 2026.