Reference to Video

A way of generating video where the model is given examples to hold to — images of a character, a product or a style, short clips for motion, sounds for the audio — alongside the prompt. Instead of inventing a subject from words, it renders the one in the references. On WowMade this is what WAN 3.0 Prime is built for: up to twelve references go in, and the clip is generated to match them.

A start frame is the literal first frame of the clip: the picture is unchanged and motion follows from it. A reference is a description by example: the model studies it and then composes a new shot that keeps the subject, style or sound, from any angle and in any framing. Use a start frame when the composition is settled; use references when the subject must reappear in shots you have not composed yet.

Up to twelve in total: six reference images, three reference videos and three reference audio clips. Video and audio references can be up to 15 seconds each and are billed as input seconds on top of the clip you generate, which is why a set of images is the cheapest way to lock a subject. The cost of your exact combination is printed on the Generate button before you run it.

It gets close and it is the most reliable method available, but generative models sample, so treat consistency as something you build rather than switch on. A reference sheet with front, three-quarter, profile and back views under the same light does most of the work; describing the subject the same way in every prompt does the rest. Generate a few, keep the best, and reuse that clip as a video reference for the next.

Only with the rights to do so. A real person's likeness needs that person's consent, and brand assets must be yours to use — uploads are screened before generation and are your responsibility under the terms of service. The Content and Safety Policy sets out what may not be created here, and every clip carries a C2PA mark identifying it as AI-generated.

A video reference carries motion and style — a walking rhythm, a camera habit, a colour grade — that stills cannot. An audio reference gives the native audio track something to follow: a tempo, an ambience, a voice quality. Both are capped at 15 seconds, and both are optional; most character work is done with images alone.

Build the subject once in the AI character generator, save it, and pull it into any prompt with an @ mention — the studio attaches its reference sheet for you. The same works for saved locations and objects, so a recurring set of references becomes a name you type rather than files you upload each time.