Seedance 2.0: the complete guide

Compared with Seedance 1.5 Pro, version 2.0 made three big changes:

1. Multimodal all-round reference. Instead of one start image, you can combine several images, video clips and audio files. Seedance understands what each reference contributes — a face, an outfit, a product, a camera move, a motion rhythm, a voice — and follows your prompt about how to combine them.

2. Unified audio-video generation. Sound is generated in the same latent space as the picture, so lip movements, footsteps and music hits line up without post-production.

3. Editing and continuation. Reference-based editing lets you change parts of an existing clip or extend a scene while keeping characters and style consistent.

ByteDance's technical report describes a dual-branch diffusion transformer built for "world complexity" — scenes with multiple subjects, physical interactions and camera movement.

Seedance 2.0 Fast runs the same multimodal approach with a lighter inference path. Expect slightly less fine detail and physics accuracy in complex scenes, but much faster turnaround and lower credit cost.

Use Fast when you are:

• testing several hooks or camera ideas for the same ad,
• producing background b-roll or social variations,
• iterating on reference combinations before a final render.

Move to Seedance 2.5 when you need longer takes (up to 30 s), 1080p output or more than 20 references.

In omni-reference mode you upload references and refer to them in the prompt. A reliable pattern:

• @image_1 — the character (clear face, neutral light)
• @image_2 — the product (clean packshot)
• @video_1 — the motion or camera move you want to borrow
• @audio_1 — optional voice line or music rhythm

Example prompt:

Vertical 9:16 UGC ad. @image_1 sits at a bright bathroom counter and picks up @image_2, turning it toward the camera. Camera follows the handheld push-in from @video_1. She says: "Three weeks in and my skin has never looked this calm." Soft room tone, bottle cap click. Natural daylight, realistic skin texture.

Keep references consistent in lighting and resolution, and avoid giving two references that conflict (for example two different faces for the same character).

Seedance 2's first/last-frame mode generates the motion between two images you provide. It is ideal for:

• Before/after reveals (messy room → clean room, bare face → finished makeup)
• Product transformations (closed box → unboxed product)
• Seamless loops and transitions between two scenes in a Reel

Tip: keep camera position similar between the two frames unless you describe the camera move explicitly in the prompt.