Vidu Q4 Reference to Video: Build Scenes from Your Own References
Give Vidu Q4 the ingredients — characters, products, places and voices — and describe the scene. It composes a new shot that keeps them all recognizable.
How Vidu Q4 Reference to Video Works
Reference to video has no starting frame. Instead of animating one picture, Vidu Q4 reads every reference you upload as a separate ingredient — a face, an outfit, a product, a room — and builds a brand-new shot that contains them. Your prompt is the script that tells it how those ingredients meet.
You point at references by their position. The first image you upload is "reference image 1", the second is "reference image 2", and voice clips are "reference audio 1" to "reference audio 3". Writing "the woman from reference image 1 hands the bottle from reference image 4 to the man from reference image 2" is how you connect people, objects and places that were never photographed together.
Because nothing fixes the first frame, you also choose the shape of the video: 16:9, 9:16, 1:1, 4:3 or 3:4. And because the model needs to know what to build, a prompt is required in this mode.
Choosing Good Reference Images
Each reference should teach the model one thing clearly. Mixed or cluttered images make it guess which part matters.
One subject per image
Crop each reference to a single person, object or location. A group photo makes it unclear which face belongs to which character.
Several angles of key faces
Front, three-quarter and profile views of the same character help the model keep that face recognizable when the camera moves around them.
Full outfits and props
A full-length shot of the costume and a separate, clean image of important props keep wardrobe and objects consistent from shot to shot.
A location plate
An empty photo of the set — a café, a street, a rooftop — gives the scene a fixed place instead of a new background on every generation.
When to Use All 12 Reference Images
Up to 12 images can go into one generation, but more is not automatically better. Use the full set when the scene has many things that must stay consistent; keep it lean when it does not.
Redundant images can work against you: five near-identical photos of the same face add nothing, while a blurry extra reference adds noise. Save the reference set for each character in a folder and reuse it in every episode — that reuse, more than the count, is what keeps a series consistent.
| Scene | Suggested references | Why |
|---|---|---|
| Simple product in a new setting | 2–4: product angles, logo close-up, location | Few elements, so extra images add little. |
| One character, one location | 4–6: three face angles, outfit, location | Enough angles to hold the face as the camera moves. |
| Two-character dialogue | 8–12: three angles per face, both outfits, the set | Each character needs its own clear set of views. |
| Recurring series episode | Up to 12: both leads, outfits, key props, two locations | Every element must match earlier episodes. |
Setting Up the 3 Voice References
Voice references let characters speak with a consistent, cloned voice and lip-synced delivery.
- Format: up to three MP3 clips, each 3–12 seconds long.
- One voice per clip. Record each speaker separately, with no music, echo or background chatter. Clear speech in a natural tone is the best starting point.
- Map voices in the prompt. Say which character speaks with which clip: "the man from reference image 2 says, 'Not tonight,' using reference audio 1."
- Keep native audio on. The voice is generated as part of the clip's audio, together with ambient sound.
- Voices are optional. With no audio references, the model still generates the scene, and native audio can still add ambient sound and effects.
Organizing a Multi-Shot Story
Vidu Q4 can cut between several shots inside one generation. The clearest way to ask for that is a numbered shot list in the prompt, with the shot size and camera for each, then the action.
Plan two to four shots for a clip of up to 16 seconds; more than that leaves each shot too short to read. Put the establishing shot first so the location is clear, keep the same references in play across shots so characters do not change, and place spoken lines in the shot where that character is on screen.
- Shot 1 — wide: establish the place and who is there.
- Shot 2 — medium: the main action or the first line of dialogue.
- Shot 3 — close-up: the reaction, the product detail or the emotional beat.
Vidu Q4 Reference to Video Output Options
These are the limits of the Vidu Q4 Preview. Longer and higher-resolution clips take more time and credits, so test at 540p and a short length first; rates are on the pricing page.
| Option | Values |
|---|---|
| Reference images | 1–12 |
| Voice references | 0–3 MP3 clips, 3–12 s each |
| Aspect ratio | 16:9, 9:16, 1:1, 4:3, 3:4 |
| Length | 3–16 seconds |
| Resolution | 540p, 720p, 1080p, 2K, 4K (10-bit at 2K and 4K) |
| Audio | Native audio on or off |
| Prompt | Required |
Reference to Video Prompt Examples
Each example names its references by position. Upload your images in the same order.
Two-character dialogue
On the rooftop from reference image 5, the woman from reference image 1 turns to the man from reference image 3 and says, "You kept it all this time?" using reference audio 1. He nods. Medium two-shot, city lights at dusk.
References 1–2: her face and outfit · 3–4: his · 5: rooftop.
Product in a lifestyle scene
The chef from reference image 1 sets the knife from reference image 2 on the wooden board from reference image 3 and slices a tomato in one smooth stroke. Close-up, warm kitchen light.
Give the product its own clean reference so its shape stays exact.
Character in a new location
The explorer from reference image 1, wearing the jacket from reference image 2, walks through the snowy forest from reference image 3 and stops to look up at the northern lights. Slow tracking shot.
A location plate keeps the forest consistent across retries.
Multi-shot chase
Shot 1: wide shot of the courier from reference image 1 sprinting down the market street from reference image 3. Shot 2: medium tracking shot as she jumps over a fruit cart. Shot 3: close-up of the package from reference image 2 in her hand. Fast, energetic.
Three shots fit comfortably in a 10–16 second clip.
Mascot narration
The robot mascot from reference image 1 waves at the camera in the store from reference image 2 and says, "Welcome back! Everything is 20% off today," using reference audio 1. Static camera, bright lighting.
One voice clip is enough for a single narrator.
Reference to Video or Image to Video?
Reference to video gives you control over what is in the scene; image to video gives you control over exactly how the first frame looks. If you already have the perfect picture and only need it to move, Vidu Q4 image to video is faster and keeps every pixel of the opening frame. If the shot does not exist yet — new combinations, new places, dialogue, a different aspect ratio — this mode is the one that can build it.
For an overview of the model itself, start at the Vidu Q4 page. When you are ready, open the Vidu Q4 generator, switch to the reference tab and upload your images in the order your prompt uses. The step-by-step Vidu Q4 guide explains every setting in both modes.
Vidu Q4 Reference to Video FAQ
How many reference images does Vidu Q4 reference to video accept?
On this site, between 1 and 12 images per generation (Vidu Q4 officially supports up to 15). Use more when several characters, outfits and locations must stay consistent, and fewer for simple scenes.
Can I use Vidu Q4 reference to video without voice references?
Yes. Voice references are optional, and you can add up to three. Without them, native audio can still add ambient sound and effects to the scene.
Is a prompt required for Vidu Q4 reference to video?
Yes. With no fixed first frame, the prompt tells the model how to combine the references, so it cannot be left empty in this mode.
Which aspect ratios does Vidu Q4 reference to video support?
Five: 16:9, 9:16, 1:1, 4:3 and 3:4. You pick one before generating.
Why does my character look different between generations?
Usually because the references are too few, too similar or too busy. Add clear front, three-quarter and profile views, crop out other people, and reuse the exact same reference set every time.
Can I use photos of real people as references?
Only with their permission. Our Terms of Service do not allow using a real person's face or voice without consent, or creating content intended to deceive.