AI Reference to Video Generator

Direct Multi-Asset Scenes with Explicit `@` Mentions

Instead of describing every face, outfit, prop, and camera move from scratch, Reference to Video lets you attach visual and audio references directly to your prompt. Type `@` in the prompt box to assign each uploaded file its exact role in the shot.

Lock Character Identity and Wardrobe

Upload a clear character portrait or turnaround sheet as `@image1` and an outfit or prop photo as `@image2`. Reference them in your prompt so the subject keeps recognizable features while performing new actions.

Place Products into Directed Environments

Pair a clean product photograph with a location reference image. Instruct the model to stage the product from `@image1` inside the lighting and atmosphere of `@image2` with controlled camera movement.

Combine Visual Assets with Camera Reference Clips

Attach a reference video clip (`@video1`) alongside your character and scene images to guide camera framing, blocking, or movement tempo without losing the identity of your uploaded subjects.

Sync Pacing with Reference Audio

Add a short voice line, beat, or sound effect in `reference_audio` so visual timing, ambient soundscapes, and character expressions align in a single generation pass.

How to Use Multimodal References in a Prompt

Assign each uploaded asset one unambiguous job so the model knows what provides appearance, environment, motion, or sound.

  1. Upload Images, Videos, or Audio

    Select a Seedance 2.0 or 2.5 model in Image to Video mode and upload your reference images, motion clips, or audio files into their dedicated slots.

  2. Tag Assets with `@` in the Prompt

    Type `@` in the prompt field to open the reference picker. Specify each asset's role—for example, 'The character in @image1 holds @image2 while walking through @image3, following the camera move in @video1.'

  3. Set Output Options and Generate

    Choose your aspect ratio, duration, and resolution, verify the live credit cost, and generate. Review the finished clip in the workspace and save or download it.

Multimodal Reference Controls

Mix image, video, and audio conditioning in one workspace without switching pipelines.

Character reference: consistent facial features and styling across dramatic lighting.

Multi-Image Conditioning

Upload multiple reference images to separate subject identity, wardrobe, key props, and background environment.

Interactive `@` Prompt Tagging

Insert `@image1`, `@video1`, or `@audio1` directly from the autocomplete menu so every uploaded file is explicitly bound to an instruction.

Video Reference Guidance

Feed short MP4 or MOV clips to guide camera path, blocking, or movement rhythm alongside your visual references.

Audio Reference & Sound Generation

Provide WAV or MP3 clips for rhythm and vocal timing, or enable native audio generation to produce synchronized sound.

Flexible Ratio & Resolution

Output in 16:9, 9:16, 1:1, 4:3, 3:4, or 21:9—or choose Adaptive ratio—across supported model resolutions.

Consistent Multi-Shot Iteration

Keep the same uploaded character and environment references while changing the action prompt to build a cohesive multi-shot sequence.

Multi-Reference Creative Examples

See how subject identity, studio lighting, and environment details come together in reference-driven scenes.

Explore related models and workflows

Sleek copper expedition rover on a black-sand beach beneath northern lights
Video model

Seedance 2.5

High-capacity multimodal referencing for longer scenes and complex camera choreography.

Explore tool
Lantern-lit wooden boat drifting through a bioluminescent mangrove canal at night
Video model

Seedance 2.0

Combine images, video clips, and audio references with joint audio-video generation.

Explore tool
Blooming white magnolia flower with water droplets on a dark slate pedestal
Video workflow

First & Last Frame to Video

Lock both the opening frame and closing target to direct smooth visual transitions.

Explore tool
Solo contemporary dancer in a crimson silk cloak mid-leap under a stage spotlight
Video workflow

Motion Transfer

Transfer choreography, action blocking, or camera movement from a reference video onto your subject.

Explore tool

Frequently Asked Questions

How multimodal references, `@` tags, and upload limits work across Seedance models.

Reference to Video is a multimodal workflow where you upload one or more images, videos, or audio files and reference them in your text prompt (using `@` tags) to control character appearance, props, environment style, camera motion, or audio timing.

After uploading files to `reference_images`, `reference_videos`, or `reference_audio`, type `@` in the prompt box. Select the file from the dropdown (such as `@image1` or `@video1`) and state its role clearly—for example, 'The woman in @image1 wears the jacket from @image2 in the cafe from @image3.'

Upload limits depend on the selected model. In the generator, Seedance 2.0 and 2.5 Mini support up to 9 images, 3 video clips, and 3 audio files (up to 15 seconds total per media type), while Seedance 2.5 supports higher multimodal reference counts and up to 30 seconds total per media type. Check the uploader hints for the active model's exact limits.

Reference images (`reference_images`) guide who or what appears in the scene and can be combined with video and audio references. First and last frames (`first_frame` / `last_frame`) lock the exact opening and closing composition of the clip. The two modes cannot be used simultaneously in the same generation.

Keep the same clean, well-lit reference image of your character in `@image1` across every generation, and only change the action, camera angle, or background reference in your prompt from shot to shot.

Yes, reference video duration and selected model parameters can affect the credit quote. The exact credit cost is calculated automatically and displayed before you click Generate.

Web search grounding is only available for pure text-to-video prompts. Once any image, video, or audio reference is attached, the generator automatically turns off web search so the model focuses on your uploaded assets.

Free sign-up trial credits apply to Seedance 2.5 Mini at 720p for standard text-to-video or single-image workflows without video or audio references. Full multimodal reference workflows with video/audio inputs and higher-tier models require a paid plan or credit pack.
Mirror-calm alpine lake at twilight reflecting snow-dusted granite peaks

Build Cohesive Scenes from Your Own Assets

Upload your character sheets, product photos, motion clips, and audio cues in one workspace. Tag each reference with `@` and preview the generated shot in your browser.