SEEDANCE 2 REFERENCE TO VIDEO

Seedance 2 Reference to Video is ByteDance's most advanced video generation model, designed to transform your creative references — images, videos, and audio — into cinematic video output with stunning realism. Whether you're a filmmaker previewing a scene, a designer animating a concept, or a content creator building engaging short-form video, Seedance 2 gives you director-level control over every aspect of the generation process.

At its core, this model excels at reference-driven video creation. Rather than generating video from text alone, you can supply up to 9 reference images, up to 3 reference videos, and up to 3 audio files to guide the output. This means you can feed in a character sketch, a mood board photo, or even a voice recording and watch the model weave those elements into a cohesive, polished video. You reference these inputs directly in your text prompt using simple tags like @Image1, @Video1, or @Audio1, giving you precise control over how each asset influences the final result.

One of Seedance 2's standout capabilities is native audio generation. The model doesn't just produce silent clips — it creates fully synchronized soundscapes including ambient sounds, sound effects, and lip-synced speech. This is enabled by default, so your videos come to life with audio that matches the on-screen action right out of the box. If you prefer to work with silent footage, you can simply toggle audio generation off.

The model produces videos with real-world physics simulation, meaning motion, gravity, fluid dynamics, and object interactions look natural and believable. Combined with its cinematic visual quality, this makes Seedance 2 particularly well-suited for narrative storytelling, product visualization, social media content, and any project where visual polish matters.

You have flexible control over the format and length of your output. Videos can be generated in durations ranging from 4 to 15 seconds, or you can let the model automatically determine the ideal length based on your prompt. Aspect ratio options are equally versatile: choose 16:9 for traditional landscape/widescreen, 9:16 for vertical content perfect for social platforms, 1:1 for square formats, 4:3 or 3:4 for classic proportions, 21:9 for ultrawide cinematic compositions, or let the model decide automatically. Resolution options include 480p for faster generation when you're iterating on ideas, and 720p for a balance of quality and speed.

The reference system is remarkably flexible. For images, supported formats include JPEG, PNG, and WebP, with each file up to 30 MB. Reference videos accept MP4 and MOV formats, with a combined duration between 2 and 15 seconds and a total size under 50 MB. Each reference video should be between roughly 480p and 720p in resolution. Audio references support MP3 and WAV formats, with up to 15 seconds of combined duration and a maximum of 15 MB per file. One important note: if you include audio references, you must also provide at least one reference image or video. The total number of reference files across all types must not exceed 12.

This multi-modal input system opens up powerful creative workflows. Imagine uploading a photo of a character, a short clip showing a specific movement style, and a voice recording — then writing a prompt that brings all three together into a seamless animated scene. The model's ability to handle stylized content and transformations makes it ideal for projects ranging from realistic live-action aesthetics to highly stylized, artistic animations.

For creators working on lip-sync projects, Seedance 2 is particularly capable. You can provide audio of dialogue or singing and reference images of a character, and the model will generate video with accurately synchronized mouth movements and expressions. This makes it a powerful tool for animation, virtual avatars, music videos, and dubbed content.

A seed value can be set for reproducibility, allowing you to regenerate similar results when refining your work. However, it's worth noting that results may still vary slightly even with the same seed, so treat it as a guide rather than a guarantee of identical output.

The model truly shines for creative professionals who want to move beyond static imagery into dynamic, story-driven content. Filmmakers can use it to pre-visualize scenes before committing to expensive production. Designers can bring product concepts to life with realistic motion. Social media creators can produce scroll-stopping vertical video content. Animators can rapidly prototype character movements and scenes. And musicians or podcasters can generate visual accompaniments to their audio content.

Seedance 2 Reference to Video represents a significant leap in AI-assisted video creation, combining multi-modal input flexibility, native audio synthesis, realistic physics, and cinematic visual quality into a single, versatile generation tool. Its ability to accept and intelligently combine text, image, video, and audio references sets it apart as one of the most comprehensive video generation models available to creative professionals today.