Vidu Q3·von ViduReference to video
Vidu Q3 Reference to Video
Vidu Q3 Reference-to-Video generates videos from 1-4 reference images with consistent subjects. Features intelligent camera switching with better consistency across multiple camera positions, audio support, and resolutions up to 1080p.
vidu/q3/reference-to-videoParameter
| Parameter | Typ | Standard | Bereich oder Optionen |
|---|---|---|---|
duration durationThe duration of the generated video in seconds. | slider | 5 | 3 – 16 |
aspect ratio aspect_ratioThe aspect ratio of the output video. | select | 16:9 | 16:9, 9:16, 3:4, 4:3, 1:1 |
resolution resolutionThe resolution of the generated video. | select | 720p | 540p, 720p, 1080p |
seed seedThe random seed to use for the generation. Set -1 for random. | number | 0 | — |
generate audio generate_audioWhether to generate audio for the video. | select | true | true, false |
Eingaben
Dateien, die dieses Modell zusätzlich zum Prompt entgegennimmt.
| Eingabe | Akzeptiert | Max. Dateien | |
|---|---|---|---|
images images | image | 4 | Erforderlich |
Prompt-Leitfaden für Vidu Q3
TARGET MODEL: Vidu Q3 (Shengshu — joint audio-video in one pass: lip-synced dialogue, voiceover, sound effects and music generated with the picture; 1–16 s; native camera control and "smart cuts" multi-shot). Plain descriptive prose. No negative prompt, no weighting syntax; the style and motion-strength parameters are documented as ineffective on Q3, so everything about look and energy is said in words.
Regeln, die gelten müssen
- Describe subject, action, setting, style, camera movement and mood as prose; the official example is one flowing paragraph ("In an ultra-realistic fashion photography style featuring light blue and pale amber tones, an astronaut in a spacesuit walks through the fog. The background consists of enchanting white and golden lights…"). No tag lists, no bracketed commands, no (word:1.3).
- Look and motion energy are words, not parameters: name the visual style in the sentence ("anime", "ultra-realistic fashion photography") and the amount of motion ("barely moves", "sprints") — the style and movement_amplitude parameters do nothing on Q3.
- Sound is generated with the picture whenever audio is on: write it as content — a spoken line in quotes attributed to a visible speaker, named sound effects tied to visible actions, the music's character. Multi-speaker conversation is supported: label each speaker uniquely and give each line its own sentence.
- Dialogue languages that are supported: English, Japanese, Chinese. Another language may render poorly — keep such a line very short or carry the beat without on-camera speech.
- One primary subject with supporting environmental detail per clip; do not name real people, brands or titles.
- Hard cap 5,000 characters; a clip prompt is a paragraph, not a script.
Was am besten funktioniert
- State the camera explicitly with film terms — pans, push-ins, tracking shots, "slow aerial drone shot", "static locked-off" — Q3 honors frame-level camera direction and a named move is more consistent than an implied one.
- Multi-shot is native ("smart cuts"): when the story has beats, write them in order, each with its framing and its sound, and let the model place the cuts; a single-take clip says so ("one continuous shot").
- Environmental sound follows environmental detail: "busy Tokyo crosswalk at night" pulls tires, train horns and signal beeps — describe the world and its sounds together ("campfire crackles… sparks drifting… crickets chirping, occasional owl hoot").
- Match content to seconds: a 5-second clip is one beat; use the 16-second ceiling for a short sequence, not for a longer single action.
Aus LUVIs eigenem Handbuch für diese Familie; dieselben Regeln wendet der Rewriter im Arbeitsbereich und die MCP-Prompt-Engine an.