Reference to video. Keep your direction.
Guide an AI video with image references for the subject, setting or style, then describe a new scene.



Whoops! Total... Intentional drift! You guys love that, don't you? Ashen? Hey, it's me. Keep this line open, yeah?
How it works
Try it now
01Add your reference
Drop it in, or pick one of yours.
- 02
Say what you want
A sentence or two is enough.
03Make it
Start creating, then keep working with HowWorks.
Explore the possibilities






About
Use reference-to-video AI when you have visual material for the scene you want to create. Add images that establish a character, an object, a place or a visual style, and explain the role of each reference in the prompt. Describe the action and camera for the new shot rather than only listing what appears in the pictures. Reference capacity and supported settings vary by model. Choose Image to Video instead when one supplied picture must be the opening frame.
Questions
What can a reference image guide?
Use references to show a subject, object, location, palette or visual treatment. State the details that matter and describe the new scene instead of asking the model to copy every image at once.
How is this different from image to video?
Image to Video uses a supplied start frame. Reference to Video uses images as visual guidance without making one the required opening. Choose the input setup according to what the shot needs.
How many references can I use?
The current model configurations support up to nine images for Seedance 2.0, MiniMax H3 and MiniMax H3 Max, and up to 30 for Seedance 2.5. Give the images clear jobs and check the selected model before starting.
Do all tiers support reference to video?
Check the selected tier. MiniMax H3 Max uses Standard for reference-to-video; its Turbo tier does not support this operation. Other settings still depend on the selected model.
Do references guarantee exact character or product details?
Inspect identifying features, proportions, materials and lettering across the actual clip. Reference support describes how the scene can be guided, not a guarantee that every detail stays identical.








