Consistent people and characters
Use portraits or a character sheet to anchor facial structure, hair, wardrobe and proportions while the person moves through new shots and locations.
Reference to video · Multimodal control · Free to try
Reference to Video AI Generator
Add character, product, style, video or audio references, then describe the scene you want. Musci3 turns those inputs into one coherent video while keeping the details that matter recognizable.
- Up to 50 references
- Native audio
- No watermark
Loading the generator…
Generation requires an account and uses credits based on the model, duration and resolution you choose. Reference limits vary by model, and failed tasks are refunded automatically.
Reference to Video Models
Seedance 2.5 is the default for large multimodal projects. Seedance 2.0, Gemini Omni Flash, HappyHorse 1.1, PixVerse v6 and Vidu Q3 offer different reference limits, durations and visual strengths.
Control More Than the Prompt
Four ways reference to video gives a written direction something concrete to follow.
Multiple references, clear roles
Combine images, source clips and audio, then state which asset controls identity, product detail, camera motion, rhythm or voice.
Products that stay recognizable
Reference the real object from several angles so its silhouette, materials, colors and packaging survive a more ambitious commercial scene.
New action without a reshoot
Keep the approved subject and art direction while changing the performance, environment, lighting and camera path in the prompt.
How to Make a Reference Video
Assign each input a job, describe the change, then let the model combine them.
Add your references
Upload the images, videos or audio clips that define the character, product, style, motion or sound. The available number and media types depend on the selected model.
Name what each one controls
Write a direct brief such as “Image 1 controls the face, Image 2 the coat, and Video 1 the circular camera move,” then describe the new scene.
Generate, compare and refine
Render a short take, check the details you locked, and regenerate with a clearer constant or change before committing to a longer version.
Where References Make the Difference
Six workflows where consistency matters as much as motion.
Consistent character stories
Carry the same original character through several shots without rewriting every facial, wardrobe and color detail in each prompt.
Product campaigns
Put one photographed product into studio, lifestyle and cinematic settings while holding its shape, finish and packaging as steady as possible.
Virtual influencers
Build short lifestyle clips around a recurring digital person, using the same identity references while outfits, locations and actions change.
Fashion lookbooks
Reference a model, garment and visual treatment separately, then direct walks, turns, close-ups and transitions for a complete fashion sequence.
Storyboard to scene
Use a storyboard for shot order and composition, a character sheet for continuity, and the prompt for timing, camera language and sound.
Branded visual worlds
Reuse approved colors, materials, sets and product photography across a family of campaign videos without recreating the art direction from scratch.
More Reference to Video Examples
Explore reference-led prompts from the Musci3 gallery. Choose an example to load its direction into the generator and adapt it to your own assets.
Reference Control, Simple Credits
Reference to video is available through the models that support it. Credit use follows the selected model and duration, with failed generations refunded automatically.
Reference to Video FAQs
What is reference to video?
Reference to video is an AI generation workflow that accepts one or more media assets alongside a written prompt. The references provide concrete information about identity, products, style, motion or sound, while the prompt explains what should happen in the new video.
How is reference to video different from image to video?
Image to video normally treats one uploaded picture as the opening frame. Reference to video can use several assets for different creative roles and does not require every image to become the first frame, making it better for identity, style and multi-shot consistency.
Which reference types can I upload?
Seedance models can combine images, video clips and audio references. Other models may accept images only or impose a smaller limit. The upload controls update when you switch models, so the generator shows the options that are actually available.
How many references can Seedance 2.5 use?
The current Seedance 2.5 configuration supports up to 30 images, 10 videos and 10 audio clips, with a maximum of 50 reference assets in one project. You rarely need that many; a small set with clearly assigned roles is easier to direct.
How do I keep a character consistent?
Use clear front, side and expression views where possible. In the prompt, state that the reference controls facial structure, hair, body proportions and wardrobe, then list only the details that are allowed to change.
Can I use product photos as references?
Yes. Product work benefits from several clean angles and close views of important materials or packaging. Tell the model which details must remain fixed, especially silhouette, label placement, color and surface finish.
Can a source video control the camera or motion?
Yes on models that accept video references. Assign the clip a specific role such as body movement, shot timing or camera path, rather than asking the model to copy everything in it. This separates useful motion from the source video's appearance.
Does reference to video generate audio?
Seedance 2.5 and Seedance 2.0 can generate native audio with the picture, and they can also accept audio references. Audio support differs by model and mode, so use the audio switch shown in the current generator settings.
What image quality works best?
Use sharp, well-lit references where the important subject is large enough to inspect. Avoid conflicting faces, heavy filters and tiny product labels. Consistent reference quality usually matters more than uploading the maximum number of files.
Why did the result ignore one of my references?
The brief may not have assigned that asset a clear role, or several references may be competing for the same detail. Name each input explicitly, reduce contradictions and put the most important identity or product constraint first.
Is reference to video free to try?
You can explore the generator and configure the project without paying. Generating requires an account and credits; new accounts receive 15 free credits, and tasks that fail are refunded automatically.
Can I use reference-generated videos commercially?
Commercial use is included on paid plans, subject to the current terms. You remain responsible for having the rights to every person, image, clip, audio recording, logo and product you upload as a reference.
Turn Your References Into a Scene
Add the character, product, style or motion you want to preserve, assign every asset a role, and direct the new video in one prompt.
Create From References