ByteDance: Seedance 1.5 Pro

Seedance 1.5 Pro is ByteDance Seed’s model for native joint audio-video generation. It models the visual action, speech and rhythm together, allowing a scene to include a character’s performance, synchronized dialogue and a designed soundscape. The central difference from a visual-only video brief is that sound becomes part of the scene description: who speaks, how the line is delivered and what the audience hears around the action.

Joint audio-visual generation

The documented sound range includes environmental sounds, action effects, synthesized sounds, instruments, background music and vocals. The model is intended to connect sound with what happens on screen: an object striking a table, a door opening or a character speaking. Use this for a short dramatic beat, a voiced product advertisement or a scene whose mood depends on both performance and sound.

Start the prompt with the subject and movement, then add the environment, camera, visual style and sound. A concise scene might be: “A chef lifts the lid from a small pot, smiles and says, ‘That is ready.’ A soft burst of steam escapes, followed by the quiet clink of the lid resting on the counter.” This gives the spoken line and effects a visible cause.

Dialogue and voiceover

The model supports monologues and conversations involving several speakers, with lip movements matched to the characters. Its guide documents Mandarin, Chinese dialects including Cantonese, Shaanxi and Sichuan, and spoken English, Japanese, Korean, Spanish and Indonesian. The documented prompt languages are Chinese and English; the language of the prompt and the language requested for a spoken line are separate choices.

Identify each speaker by a consistent visible feature and write the lines in quotation marks. Include delivery instructions such as a calm voice, a hesitant response or a faster pace, and make turn-taking explicit. For two speakers, describe who speaks first and whose reaction the camera shows. A short exchange that fits the chosen clip gives the model a clearer performance target than a long passage with unspecified speakers.

Voiceover can be directed through timbre, emotion, intonation and speech rate. It is useful when the voice should explain a product while the camera stays on the product, or when a narrator sets the mood without an on-screen speaker. The guide describes maintaining a recognizable timbre while changing delivery; review the actual wording, pronunciation and performance together with the image.

Sound effects and background music

Describe effects as events connected to the action: footsteps as a character enters, rain becoming stronger outside a window or fabric moving as a coat is put on. Background music is generated to match the prompt by default. Specify its mood, style and rhythm when those matter, and state when it should begin or fade relative to the scene.

For a quiet demonstration, request restrained music and emphasize the product’s handling sounds. For an animated scene, describe the action beat that should coincide with the musical rhythm. The generated soundtrack is mono. The documented audio control also allows a silent video when the project needs a separately prepared soundtrack.

Shot transitions and camera direction

The model supports shot changes with consistent visual treatment, reverse shots for conversation and instructions about transition timing. A useful dialogue sequence can establish two speakers in a medium shot, cut to the responding speaker and return to the first speaker’s reaction. State the identity of each person when a shot changes.

The guide covers high, low and eye-level angles; over-the-shoulder and subjective viewpoints; and wide, medium and close framing. Describe a camera move through its starting composition, direction, amplitude and ending composition. For example, a slow dolly-in can end on the speaker’s expression as a line finishes. Connect the movement to the story rather than listing several camera techniques without a purpose.

Modes, output and preview

The documented input modes are text-to-video, first-frame image-to-video and first-and-last-frame image-to-video. A prepared first image establishes the starting composition; a final image can define the intended endpoint. The output specifications list MP4 at 24 fps, 480p, 720p or 1080p, and 4–12-second duration settings. The model can also choose a whole-second length within its supported range. The documented ratios are 21:9, 16:9, 4:3, 1:1, 3:4 and 9:16, with adaptive framing supported.

The manufacturer documents a two-stage Draft workflow: a 480p preview to inspect scene structure, shot choices, motion and prompt intent, followed by a final generation that reuses the creative inputs. This is a way to judge a planned performance before choosing its final resolution. The preview itself is limited to 480p and does not support returning the last frame. The producer’s Draft task ID remains valid for seven days after creation; generate the final video from it within that window.

For image-led work, prepare the source picture for the intended framing: changing the output ratio can crop the image. A repeated seed produces similar rather than guaranteed identical frames, and a fixed-camera instruction is not a perfect lock guarantee. Most importantly, judge a dialogue clip as one audio-visual performance: the correct speaker, audible line, facial reaction and camera emphasis all need to support the same event.