Skip to content
Kenerate IconKenerate Icon
Kenerate AI
Start

23 reference models · Wan 3 · Seedance 2.5 · MiniMax H3

Reference to Video AI — the same character in every shot

Upload images of your character, mascot or product, then describe a new scene — the AI keeps them looking the same in every shot. Up to 30 reference images on Seedance 2.5, and images, videos and audio on Wan 3.

Try
Looping clip of the explorer character in a spice market beside her reference photo
  • Consistent characters
  • Products stay on-brand
  • Up to 30 ref images
  • Videos and audio too (Wan 3)

Updated by the

Made with Kenerate

Quick answer

Updated

What is reference to video AI?

Reference to video AI takes several reference images, like a face, an outfit or a product, and generates a new scene that keeps them looking the same, for a series of shots with one consistent character. In Kenerate AI's studio 23 models support it; Seedance 2.5 takes up to 30 images and Wan 3 takes images, videos and audio.

Key facts

Reference models
23
Most ref images
30 · Seedance 2.5
Wan 3 refs
10 img · 5 video · 5 audio
Longest clip
30 s
Up to 9 images
MiniMax H3 Pro, Ken 1
Veo 3.1
Up to 3 images

Real results

One character, many shots

Real Wan 3 reference-to-video clips (5 s, 720p, sound). The same explorer, robot and sneaker — from three Seedream 5.0 Pro reference images.

Real Wan 3 reference-to-video clips (5 s, 720p, with sound). Open one to see its reference images and prompt.

How to

How to use Kenerate Reference to Video AI

Time needed
Steps
4 steps

  1. Step 1

    Make clean reference images

    One subject per image on a plain background — like the explorer, robot and sneaker here, made with Seedream 5.0 Pro.

  2. Step 2

    Open the video studio in reference mode

    Pick Wan 3, Seedance 2.5, MiniMax H3 Pro or Ken 1 and add your references.

  3. Step 3

    Name them in the prompt

    On Wan 3, refer to uploads by order: 'Image 1 kneels and high-fives Image 2'.

  4. Step 4

    Generate every shot the same way

    Reuse the same references for each scene to keep the character consistent.

1Describe the scene2Or start from an image3Pick a video model4Set ratio and length, then Create5Watch and download
Kenerate AI Video Studio in Text to Video mode with a lighthouse scene prompt typed in

Open Video Studio in Text to Video mode and write the shot — subject, action, camera move and light.

1 / 5 · Describe the scene

Watch the tour · 1:19

Text, image & reference to video, models, edit, extend & chat

Chapters

Engines

Models that take references

From the studio's own model list, with how many references each takes.

  • Wan 3 title card — a long cinematic clip with sound
    Used here

    Alibaba

    Wan 3

    Input
    Up to 10 images, 5 videos, 5 audio clips
    Output
    2–30 s · up to 1080p · sound

    Best for

    Most reference types at once

    Open Wan 3
  • Seedance 2.5 title card — a multi-shot cinematic triptych of a heist sequence in crisp 4K

    ByteDance

    Seedance 2.5

    Input
    Up to 30 images, 3 videos, 10 audio clips
    Output
    4–30 s · up to 4K · sound

    Best for

    The most reference images

    Open Seedance 2.5
  • Seedance 2.0 (Fast, Mini, Turbo) title card — three film strips of the same cinematic sunflower-field scene

    ByteDance

    Seedance 2.0 (Fast, Mini, Turbo)

    Input
    Up to 9 images, 3 videos, 3 audio clips
    Output
    4–15 s · up to 4K · sound

    Best for

    Seedance 2 quality

    Open Seedance 2.0 (Fast, Mini, Turbo)
  • MiniMax H3 Pro title card — reference photos of a character and a location merging into a cinematic film still

    MiniMax

    MiniMax H3 Pro

    Input
    Up to 9 images, 3 videos, 3 audio clips
    Output
    4–15 s · 768p or 2K

    Best for

    2K reference clips

    Open MiniMax H3 Pro
  • Ken 1 title card — a bold cinematic film still of a lone fictional adult rider on a motorbike across a neon salt flat at dusk

    Kenerate AI

    Ken 1

    Input
    Up to 9 images, 3 videos, 3 audio clips
    Output
    3–15 s · up to 1080p

    Best for

    Kenerate's own model

    Open Ken 1
  • Veo 3.1 title card — two fictional adult characters in dialogue on a rainy noir street

    Google

    Veo 3.1

    Input
    Up to 3 images
    Output
    8 s · up to 1080p · sound

    Best for

    Google's 'ingredients to video'

    Open Veo 3.1
All models (10)
  • PixVerse v6 title card — a stylised social clip of a neon city at night with motion streaks

    PixVerse

    PixVerse v6

    Input
    Up to 10 images, 2 videos
    Output
    1–15 s · up to 1080p

    Best for

    Stylised multi-reference clips

    Open PixVerse v6
  • Grok Imagine 1.5 title card — a punchy social clip still of confetti exploding at a rooftop party of fictional adults

    xAI

    Grok Imagine 1.5

    Input
    Up to 7 images
    Output
    1–15 s · up to 720p

    Best for

    Quick reference clips

    Open Grok Imagine 1.5
  • Happy Horse 1.1 title card — a cheerful everyday clip still of a horse galloping through a sunny meadow

    Alibaba

    Happy Horse 1.1

    Input
    Up to 9 images
    Output
    3–15 s · 720p or 1080p

    Best for

    Everyday reference clips

    Open Happy Horse 1.1
  • Vidu Q3 title card — reference images of a fictional adult character and a costume combining into a cinematic scene

    Vidu

    Vidu Q3

    Input
    Up to 4 images
    Output
    1–16 s · up to 1080p · sound

    Best for

    Vidu's reference mode

    Open Vidu Q3

The examples above were made with Wan 3 reference to video.

Tips

Consistency that holds

From Kling's, Alibaba's and ByteDance's guides, and our tests.

  • One subject per reference image, plain background.

  • Add more angles of the same character for turns and close-ups.

  • Refer to references by order on Wan 3: Image 1, Image 2.

  • Keep the outfit in the reference if it should stay.

  • Reuse the exact same references for every shot in a series.

  • For a voice, add a reference audio clip on Wan 3 or Seedance.

Example uses

How creators use reference to video AI

Illustrative examples of typical workflows, not customer reviews.

  • The same sneaker in a studio, on a street and on a trail

    Brand marketer

    A marketer can reuse one clean product reference so a sneaker looks identical in every ad shot.

  • Small 3D robot mascot on a sunny kitchen counter

    Mascot designer

    A mascot designer can keep a 3D character identical across a series of short scenes.

  • Explorer seen from behind on a snowy ridge at sunrise

    Indie filmmaker

    A filmmaker can keep the same lead character across shots, from a market to a snowy ridge.

FAQ

Reference to Video AI questions

What is reference to video in AI?

A mode where you upload images (and on some models videos and audio) of a character or product, and the model puts them into a new scene while keeping them looking the same.

Link to this answer

What's the difference from image to video?

Image to video keeps your picture as the first frame. Reference to video uses the images as a guide for who or what appears, then builds a new scene around them.

Link to this answer

How do I keep the same character in every shot?

Use the same reference images for every clip, one subject per image on a plain background, and name the subject the same way in each prompt.

Link to this answer

How many reference images can I use?

Up to 30 on Seedance 2.5, 10 on Wan 3 (plus 5 videos and 5 audio clips), 9 on MiniMax H3 Pro, Ken 1 and Happy Horse 1.1, 3 on Veo 3.1.

Link to this answer

Can I put two characters in one clip?

Yes — add both references and describe what each does, like the explorer and robot example: 'Image 1 kneels and high-fives Image 2'.

Link to this answer

Can I keep a product on-brand across videos?

Yes — use a clean product reference, like the sneaker here, and reuse it in every ad.

Link to this answer

Is reference to video free?

Video generation runs on one-time credit packs that never expire. The free sign-up credits work on image models, so you can make the reference images first.

Link to this answer

In depth

Reference to Video AI, explained

Why references keep characters consistent

A single start image shows one angle, so the model has to guess the rest as soon as the camera moves. Several references tell it who the character is from every side. Kling puts it simply: one image shows what a character looks like from one angle; several show what the character actually is.

Explore

Every Kenerate video tool

Generators, tools, models and guides — all in one studio.

See more of Kenerate AI in your Google resultsAdd kenerateai.com as a preferred source — Google shows more from your preferred sources in Top Stories and AI answers.

Keep going

More from Kenerate

By Last updated and reviewed

Sources: Kenerate video studio · ByteDance Seed — Seedance 2.5 · Alibaba — Wan prompt guide · Kling — Element library guide · Vidu — reference to video

Try Kenerate Reference to Video AI

One studio for images and video.

Try reference to video

Tried it? Tell us how it went