All-purpose Reference to Video: AI from Images, Clips & Audio
Combine reference images, video clips and audio with a prompt to generate one AI video. The upload area shows only the inputs the selected model supports.
Supported models: Seedance 2.5 · Seedance 2.0 · Seedance 2.0 Fast · Seedance 2.0 Mini · MiniMax H3 · Wan 3.0 · Wan 3.0 Prime · Kling v3 · Kling v3 Omni · Veo 3 · Veo 3.1 Fast · Gemini Omni Flash · Grok Video 1.5 · Grok Video 1.0 · Wan 2.6 · Wan 2.6 Flash · HappyHorse 1.0 · HappyHorse 1.1
What is all-purpose reference to video?
All-purpose reference to video generates one AI video from a text prompt plus a mix of references: images for characters and style, video clips for motion, and audio for rhythm or voice. Text to video starts from words and image to video animates a single picture, while reference to video lets you steer several aspects of the result at once.
Each model accepts different references. Wan 3.0 takes up to 21 references across images, clips and audio; MiniMax H3 accepts up to 9 images, 3 video clips and 3 audio files; and HappyHorse 1.0 uses up to 9 reference images. The upload area adapts automatically, so you only see the slots the selected model can use.
What you can do with reference to video
Consistent characters
Keep the same face, outfit and props across shots by referencing character images.
Copy motion from a clip
Use a reference video to guide movement, choreography or camera work.
Music-driven visuals
Add an audio reference so the pacing and mood follow the track.
Product videos from real assets
Combine product photos with a lifestyle scene reference for on-brand ads.
Style transfer for video
Apply the look of an artwork or brand image to a new scene.
From storyboard to shot
Turn storyboard frames and a rough clip into a finished sequence.
AI models for reference to video
Reference limits differ by model. Output lists the resolution range and clip lengths.
| Model | Best for | Output |
|---|---|---|
| Wan 3.0 | The widest mix: up to 21 image, video and audio references, with clips up to 30 seconds | 480p–1080p · 2–30s |
| MiniMax H3 | Images, clips and audio together: up to 9 images, 3 clips and 3 audio files | 480p–4K · 4–15s |
| HappyHorse 1.0 | Multi-image character references with up to 9 images | 720p–1080p · 3–15s |
| Veo 3.1 Fast | Up to 3 reference images with generated audio | 720p–4K · 4–8s |
| Wan 2.6 | Reference video clips that guide motion | 720p–1080p · 5–10s |
How to create a video from references
- 1
Choose a model
Pick a video model. The reference slots switch to the images, clips and audio that model accepts.
- 2
Add references and a prompt
Upload your references, then describe how each one should shape the video. Where supported, type @ to mention an added image in the prompt.
- 3
Generate and download
Check the estimated credits, generate, and download the finished video from your history.
Tips for working with references
- Give every reference a job in the prompt, for example: use image 1 for the character and the clip for the dance moves.
- Use clean references: a front-facing portrait or character sheet works better than a busy group photo.
- Keep reference clips short and focused on the one motion you want to copy.
- Use audio references for pacing and mood, and describe the visuals you want in the prompt.
- Start with a few references, then add more once the base result looks right.
Use image 1 as the main character and the reference clip for the dance moves; neon-lit rooftop at night, handheld camera, vertical video
More AI tools to try
- AI Image to VideoAnimate a photo with AI image to video. Upload an image, describe the motion, and generate with Wan 3.0 Prime, Kling v3, HappyHorse or Veo 3.1 Fast.
- AI Text to VideoCreate AI videos from a text prompt. Compare MiniMax H3, Wan 3.0, Kling v3 and Veo 3.1 Fast, set length and resolution, and see the credits before you generate.
- AI Video EditorEdit videos online with AI: restyle, colour-grade and remix footage in minutes, with no software to install.
Frequently asked questions
- What is all-purpose reference to video?
- It generates one AI video from a text prompt plus reference images, video clips and audio, giving you more control over characters, motion and mood than text or a single image alone.
- How is it different from image to video?
- Image to video animates a single picture and accepts images only. Reference to video combines several references, including images, clips and audio, to steer a new scene.
- Which references can I add, and how many?
- It depends on the model. Wan 3.0 accepts up to 21 references in total, MiniMax H3 up to 9 images, 3 video clips and 3 audio files, and HappyHorse 1.0 up to 9 images. The upload area only shows what the selected model supports.
- What file formats are supported?
- Images in common formats such as JPG, PNG and WebP, video clips in MP4, MOV or M4V, and audio files such as MP3.
- How many credits does it cost?
- The cost depends on the model, resolution and length, and on some models the length of reference clips adds to the billed duration. The estimate is shown before you generate.
- How long does generation take?
- Usually from under a minute to several minutes, depending on the model, the references and the length. It runs in the background, so you can check your history later.
- Do the videos have a watermark?
- Watermark-free downloads are included with paid plans. See the pricing page for what each plan includes.
- Can I keep the same character across several videos?
- Yes. Reuse the same character images as references in each generation and describe the character consistently in your prompts.