Each clip starts from reference images, a reference video or a voice line, with the references shown along the bottom. Hit Try this example to load the same references and prompt into the generator.
Reference to video turns the people, objects, places, motion and sound you upload into a new AI video, instead of animating one fixed photo.
With reference to video AI, every file you upload is a reference, not a first frame. Add a photo of a person, a product shot and a location, write what should happen, and the model casts all of them into one new scene. Reference images alone are enough: reference image to video is the most common way to use it, and it keeps faces, outfits, logos and colors consistent from shot to shot.
Add a short video clip and the model follows its motion, camera move and timing, while your images decide who and what appears on screen. Add an audio clip and it becomes the voice line, music or sound of the new video. Seedance 2.0, MiniMax H3 and Wan 3.0 take all three in one generation.
People, characters, products, pets, outfits or places. Name them in your prompt as @Image 1, @Image 2 so the model knows who is who.
Up to 9 images · JPG, PNG or WebP
A dance, a walk, a camera move or a whole shot to copy. The model keeps the motion and framing and swaps in the people and objects from your images.
Up to 3 clips, 15 s in total on most models · MP4, MOV or WebM
A voice line, a music loop or a sound effect becomes the soundtrack of the new clip. On Seedance 2.0, a voice line is spoken by the person on screen with matching lip movement.
Up to 3 clips, 15 s in total · no extra credits
All three modes live in the same generator. Pick by what you upload and what has to stay fixed.
| Reference to Video | Image to Video | Start & End Frame | |
|---|---|---|---|
| What you upload | 1–9 images, plus optional video and audio clips | 1 image | 2 images: the first and last frame |
| What the images do | Tell the model who and what to show; the scene is composed from scratch | The photo becomes the first frame; the scene stays as photographed | Fix how the clip starts and ends |
| Where the motion comes from | Your prompt, or a reference video | Your prompt | Your prompt, between two fixed frames |
| Best for | The same character across shots, product ads, putting people into new scenes, copying a dance or camera move | Bringing one photo to life | Transitions, before and after, morphs |
| Sound | Generated, or taken from your reference audio | Generated on most models | Generated on most models |
Only have one photo? Use Image to Video. Need a clip that starts and ends on exact images? Use Start & End Frame.
Every model below works with reference images. Seedance 2.0, MiniMax H3 and Wan 3.0 also take reference video and audio.
| Model | Images | Reference video | Reference audio | From (credits) |
|---|---|---|---|---|
| Seedance 2.0 Mini | Up to 9 | 3 clips · 15 s | 3 clips · 15 s | 15480p · 4s |
| Veo 3.1 Lite | Up to 3 | — | — | 15720p · 8s |
| Veo 3.1 Fast | Up to 3 | — | — | 30720p · 8s |
| Grok Imagine 1.5 | Up to 7 | — | — | 2480p · 1s |
| MiniMax H3 | Up to 5 | 3 clips · 15 s | 3 clips · 15 s | 32720p · 4s |
| Wan 3.0 | Up to 9 | 3 clips · 15 s | 3 clips · 15 s | 16480p · 2s |
| Wan 3.0 Prime | Up to 9 | 3 clips · 15 s | 3 clips · 15 s | 24480p · 2s |
| Gemini Omni Flash | Up to 7 | — | — | 451080p · 4s |
| Seedance 2.0 Fast | Up to 9 | 3 clips · 15 s | 3 clips · 15 s | 47480p · 4s |
| Kling 3.0 | Up to 7 | — | — | 54720p · 3s |
| Seedance 2.0 | Up to 9 | 3 clips · 15 s | 3 clips · 15 s | 76480p · 4s |
| Seedance 2.5 | Up to 9 | 3 clips · 30 s | 3 clips · 30 s | 112480p · 4s |
"From" is the lowest price with reference images only. A reference video adds credits for every second of the clip you upload; reference audio is free. Once you add a video or audio, models that can't use it are greyed out.
Drop in up to 9 images of the people, products and places you want in the video. Add a short video clip to copy its motion, or an audio clip for the sound.
Write what happens and point to each reference by number, like @Image 1 or @Video 1. Then pick a model, aspect ratio and length.
Click Create. Most clips are ready in a few minutes. Download the MP4, or rerun the same references on a stronger model.
Keep the same face and outfit across a series of clips for stories, comics and recurring mascots.
Combine a product photo, a model and a setting into an ad that keeps the logo and colors right.
Drop a person or a pet into a new place and let them act it out, from red carpets to rooftop chases.
Upload a reference video and swap in your own character while the moves, framing and pacing stay the same.
Add a voice line, a jingle or a beat and get a video whose soundtrack is your audio, with your character speaking your exact words.
Upload character sheets, a location and a storyboard frame, and turn them into a multi-shot sequence.
Most bad results come from the model not knowing which reference is which. A clear prompt fixes most of them.
Number your references in the order you upload them: @Image 1, @Image 2, @Video 1.
Give each one a job: "@Image 1 is the main character", "@Image 3 is the room".
With a reference video, say what to keep (motion, camera, timing) and what to replace (the person, the outfit, the setting).
Describe the action and camera in plain words. For several shots, write them as Shot 1, Shot 2.
@Image 1 is the main character. She walks her dog @Image 2 along the beach @Image 3 at sunset. Follow the camera move and pacing of @Video 1. Keep her face and outfit exactly as in @Image 1.
Uploading references and writing prompts is free. The cheapest clip is Seedance 2.0 Mini at 480p: 15 credits for 4 seconds from reference images, exactly what you get for signing up. A reference video adds credits for each second of the uploaded clip; reference audio adds nothing. 720p needs a Starter plan, and 1080p and 4K need Pro.
Credits
Features
Credits
Features
Credits
Features
Credits
Features
Upload images, a video clip or audio, describe the scene, and get a new video with your characters kept consistent.
Compare Plans