Seedance guide: references, multi shot scenes and style

What sets Seedance apart

From the start Seedance focused on clips made of several connected shots that keep the subject, style and atmosphere across cuts. Seedance 2.0 added text, image, audio and video inputs and joint generation of picture and sound.

The Seedance 2.5 guide allows up to 50 assets per task: up to 30 images, up to 10 videos totaling 30 seconds and up to 10 audio clips totaling 30 seconds. The model is most stable with 1 to 8 subjects in images and with video clips of 5 to 10 seconds.

Reference roles

Assets are named by type and number: image 1, video 1, audio 1. For each one, say what to take from it (face, costume, motion, voice) and what to leave out (background, other people). Captions drawn inside an image are not read as instructions.

If an image is the first or last frame, say so in its own sentence at the start of the prompt. Upload different angles of one object as separate images and state that they show the same object.

• "Image 1 is the heroine, use her face and outfit. Image 2 is the bag, use its shape and color. Video 1 provides only the camera move, do not carry over its people or background"
• "Image 1 is the first frame. Image 2 is the last frame. The heroine walks across the hall and sits by the window"

Multi shot scenes and timestamps

Split a long clip into stages. Each stage has a starting state, one main event and a visible outcome. Add a consistency block that fixes who wears what, who holds which prop, the screen direction and the sound.

Timestamps use one second steps, and ranges must be continuous with no overlap. ByteDance stresses that timestamps allocate pacing rather than frame accurate cuts.

• "0 to 6 s: wide shot, a courier steps out of the elevator carrying a box"
• "6 to 12 s: medium shot, he rings the doorbell and the door opens"
• "12 to 18 s: close up, a young woman takes the box and smiles"
• "Consistency: courier in a yellow jacket, box with a blue ribbon, stairwell ambience"

Camera and style

Seedance understands shot sizes from extreme wide to extreme close up, camera moves (push, pull, pan, tilt, track, orbit, crane, handheld) and low, high and point of view angles. For an unusual term, describe the visible result: what is in frame and where and how fast the camera moves.

Style comes from light, color, materials and texture. Convey emotion through two to four observable cues such as gaze, brows, breath or hands rather than a single word like "sad".

Sound and speech

The guide defines markers for audio: music in round brackets, sound effects in angle brackets, dialogue in curly braces and subtitles in square brackets. For a line, give the language, accent, delivery and speaker, then the words. State separately what to keep and what to drop, for example "no background music, no subtitles".

ByteDance recommends adding exact subtitles, signage, formulas and product specs in post production. The model tries to render them, but letter perfect text is not guaranteed.

Editing and extension

Seedance 2.5 Edit changes an existing clip from a text instruction, and the aspect ratio and duration follow the source video. Extension also inherits the aspect ratio, while the duration is yours to set. Keep boundary frames, motion and audio continuous.

How to do it in Solistra

The Solistra catalog includes Seedance 2.5, Seedance 2.5 Edit, Seedance 2.0 and faster versions. Seedance 2.5 generates 4 to 30 seconds at up to 1080p.

• Open Video and choose Seedance
• Upload your assets and state each role in the prompt
• Describe the story stages, camera and sound, then pick duration and format
• Check the credit price on the button and run it