Wan 3.0 Prime Reference to Video for Consistent Scenes
Last verified: October 29, 2026
Wan 3.0 Prime works best when you build a scene from material you already have, such as a character, a product, a location or a voice, and need each piece to stay recognizable through a clip of up to 30 seconds. With Wan 3.0 Prime Reference to Video, the images, videos and audio you upload act as reference material. The model reads your prompt to work out what each one should contribute. It then generates a new video that keeps the referenced subject's appearance, and also its voice when the reference includes one.
Prime is the speed-optimized tier of Wan 3.0. It has the same feature set as the standard model and generates much faster without losing quality. That speed matters most in reference work. Plan on a few quick drafts to get the face, the product and any borrowed motion working together before you render the version you keep.
On Vidofy you start with images: at least one goes into Add Image. Add video or audio references only when motion or sound should follow a real example. The prompt does the directing. It says who appears, what each reference contributes and what the audience hears. Video Duration, Resolution, Aspect Ratio and Generate Audio then shape the clip you get back.
Explore more reference to video
Inputs, Length and Output at a Glance
A quick summary of what the Wan 3.0 Prime form on Vidofy accepts and returns.
Input
Up to 10 images (JPG, JPEG, PNG or WEBP, up to 10 MB each) + up to 5 videos (MP4 or MOV, up to 100 MB each) (optional) + up to 5 audio files (WAV or MP3, up to 25 MB each) (optional) + a text prompt of up to 5,000 characters
Video Duration
3 to 30 seconds
Resolution
480P, 720P or 1080P
Aspect Ratio
Auto, 1:1, 3:4, 4:3, 9:16, 16:9
Audio
Switch on or off, on by default
Output
1 video per run
Prompting Techniques for Reference Videos
Naming References as Image 1, Video 1 and Audio 1
Scripting Long Clips as Timed Shots
Matching a Voice with an Audio Reference
How to Make a Reference Video with Wan 3.0 Prime
Six steps, from the first reference image to the downloaded clip.
Upload Your References in "Add Image"
Put at least one image into "Add Image". It's the only required upload. Give each subject its own clean picture, such as one for the character, one for the product and one for the location, instead of one busy collage, so the prompt can point at each one separately. Upload them in the order you plan to name them in the prompt (see the tip on naming references).
Add Motion or Sound in "Add Video" and "Add Audio"
Both are optional. Use "Add Video" when a real clip should lend its camera move, action or effect, and "Add Audio" when a voice or a music track should shape the soundtrack. Trim each reference clip to the moment you actually need. The troubleshooting section explains how reference video length interacts with the duration you pick.
Describe the Scene in the "Prompt" Box
The "Prompt" box is required and holds up to 5,000 characters, which is enough for a full shot-by-shot script. Say who each reference is, what happens, how the camera moves and what is heard. If you only have a rough idea, click "Enhance Prompt" under the box for fuller suggestions. Your text stays unchanged until you pick one.
Set "Video Duration", "Resolution" and "Aspect Ratio"
"Video Duration" runs from 3 to 30 seconds in one-second steps and starts at 3, which is too short for most scenes with dialogue. Raise it to fit the action you wrote. "Resolution" offers 480P, 720P and 1080P, with 480P preselected. Stay at 480P while you check that your references hold, then switch to 1080P for the version you publish. "Aspect Ratio" defaults to 1:1. The other options are 9:16 for vertical feeds, 16:9 for widescreen, 3:4 or 4:3, and Auto. "Generate Audio" is on by default; leave it on for dialogue and sound effects. "Seed" takes a number, and keeping it fixed makes reruns easier to compare while you change one setting at a time.
Start the Run with "Generate"
Click "Generate" once everything is set. Each run returns one video, so between runs change a single thing (a reference, a line of the prompt or a setting) to see clearly what made the difference.
Review and Download the Result
Before you download, check that each reference held: the face and outfit, the product's shape and label, the setting and, if you uploaded audio, the voice. If one element drifted, tighten how the prompt names that reference instead of rebuilding the whole setup.
Wan 3.0 Prime or Standard Wan 3.0 for Reference Video?
How the speed-optimized Prime tier compares with the standard Wan 3.0 model for the same reference task.
| Criterion | Wan 3.0 Prime | Alternatives | Best For |
|---|---|---|---|
| Turnaround while iterating | It is the speed-optimized tier of Wan 3.0, documented as much faster while keeping high quality. | Standard Wan 3.0 is the regular-speed model that Prime's faster generation is measured against. | Pick Prime when you expect several drafts per scene, because speed is where the documentation separates the two tiers. |
| Reference inputs | It supports all-in-one reference generation from multiple reference images, videos and audio in a single run. | Standard Wan 3.0 has the same documented all-in-one reference support for images, videos and audio. | Reference needs alone won't settle the choice, because both read the same mix of inputs, so let turnaround decide. |
| Output range | It shares the standard model's resolution tiers and duration range and outputs 30 fps MP4. | Standard Wan 3.0 lists the same resolution tiers, the same duration range and the same 30 fps MP4 output. | The output specs give no reason to switch tiers, so your full-length, full-resolution final can stay on Prime. |
Fixing Common Reference Video Problems
Common problems with reference videos and how to fix them.
The video won't generate when I add a reference clip and pick a long duration
My reference video gets rejected even though it's an MP4
My whole storyboard grid shows up as a picture in the video
There's music or speech in my clip that I never asked for
Frequently Asked Questions
Do I need reference videos or audio, or are images enough?
Images are enough. Add Image is the only required upload, and a character, a product and a location as separate images are already enough to stage a scene. Add Video and Add Audio are for when motion, a camera move, a voice or a music track should follow a real example. The model reads all three kinds of reference in the same run, so you can add them once the look is settled.
Which files can I upload, and what are the size and length limits?
Add Image is required and takes 1 to 10 images in JPG, JPEG, PNG or WEBP, each up to 10 MB and at least 300×300 pixels. Add Video is optional and takes up to 5 clips in MP4 or MOV, each up to 100 MB, with 15 seconds of video in total across all clips. Add Audio is also optional and takes up to 5 files in WAV or MP3, each up to 25 MB, again with 15 seconds in total. Because the 15 seconds is shared across files, one long clip can use the whole allowance. Short, trimmed clips let you take motion or sound from more than one source.
What makes one generation cost more than another?
The credit cost of a run is calculated from the options you choose and shown above the Generate button once you're signed in. It changes with Video Duration, Resolution and the length of any reference videos you upload. Check the figure after changing any of these, especially before a long final render.
Can I replace a person in a video with my own character?
Yes, swapping a character into existing footage is one of the documented reference tasks. Put the source clip in Add Video and the new character in Add Image, then name both in the prompt. For example: replace the woman in Video 1 with the man from Image 1, followed by what he does. If the original movement should stay, say so. The documented example swaps a skateboarder for a new character while keeping the action, the trajectory and the scene unchanged.
Can Wan 3.0 Prime follow a sequence of keyframe images?
Yes. Keyframe reference is a documented task: several keyframe images plus a prompt produce a video that follows the order of those keyframes. Upload them in Add Image in the order they should appear, describe in the prompt what happens between each one, and pick a Video Duration long enough for every transition.
Can I turn a clay render into a photorealistic video?
Yes, re-rendering a clay render video as photorealistic footage is a documented reference task. Put the clay render in Add Video and add at least one image in Add Image showing the characters, materials or set the final shot should have. Then describe surfaces, lighting and lens in the prompt so the realistic version has a clear target.