Seedance 2.5 Prompt Guide for Music Videos: Formula and Prep (2026)

By the AI Music Video GuideUpdated October 30, 20267 min read

Quick answer: Write a Seedance 2.5 prompt as a short director's brief in Dreamina's order: format, subject, action, environment, camera, look, timing, audio and constraints. Tag every reference with a role, give each shot one camera move, time the beats at 6 to 8 seconds, and storyboard the whole song into 30-second blocks before you spend a credit.

Seedance 2.5 changed what a prompt has to do. Dreamina's official guide, published August 18, 2026, says a 2.5 prompt steers a full 30-second clip rather than a 5-second experiment, with up to 50 reference assets across text, images, video and audio. This guide is for musicians rendering a music video with it. We work from three guides and label the version each covers: Dreamina's covers Seedance 2.5; InVideo's (June 9, 2026) covers 2.0; Atlabs' (undated, read October 25, 2026) covers 1.0 Pro. The older two still hold for vocabulary, not for limits.

What is the Seedance 2.5 prompt formula?

Dreamina's eight-part order is format, subject, action, environment, camera, look, timing, then audio and constraints. Put the facts that cannot change first, describe motion and timing next, and finish with sound and the things the model must not do. InVideo's 2.0 formula and Atlabs' 1.0 Pro syntax are shorter versions of the same order, and all three put subject and action before style.

Read the example column top to bottom and you have the finished prompt; we wrote it as a model of the structure, not a documented output. InVideo's 2.0 guide adds two mechanics: the first 20 to 30 words carry the most weight, because the model locks subject and core action from the opening, and 2.0 had a hard cap of 3,000 characters per prompt. Dreamina publishes no cap for 2.5 and says to use the shortest prompt that fully specifies the production. Atlabs' 1.0 Pro figures, 3 to 12 second clips at 480p to 1080p, are historical; OiiOii's launch release of August 7, 2026 describes 2.5 as a 30-second single take, double the 2.0 length, with up to 50 references. PrompTessor (reviewed September 5, 2026) adds that duration and audio features depend on the partner platform, so check the duration menu where you render.

How do you use references in a music video prompt?

Give every uploaded asset a named role inside the prompt. InVideo's 2.0 rule is that text defines the world, images lock identity, video guides movement, and audio shapes rhythm and sound. Dreamina's 2.5 guide says to tag references with handles such as @Image1 and bind them to specific beats, and to attach two or three portrait stills per lead so faces stay stable across shots.

InVideo's 2.0 limits show why the whole song never goes in as one input: 9 image references, 3 video references totalling 15 seconds or less, 3 audio references totalling 15 seconds or less, 12 files in total. Dreamina says 2.5 raises the pool to 50 assets but lists no per-file audio length. On either version, cut the slice of the track that matches the clip and feed that, not the master. If the audio reference misbehaves, our Seedance reference audio troubleshooting covers the common causes.

PrompTessor says to name each supplied asset and state its exact role, and never to reference an asset you did not upload. InVideo says a prompt should get shorter once a reference frame carries the look, and warns that you need the rights to anything you upload: your own track, a consenting performer, no third-party logos. For the performer, lock one primary portrait and repeat its fixed traits at each beat; the character consistency guide goes deeper.

What shot and camera language does Seedance follow?

Name the shot size and one camera move, and put them before lighting and mood. Dreamina's troubleshooting table says a camera move gets ignored when it is buried under style language, and prescribes one move per shot. Atlabs' 1.0 Pro guide supplies the vocabulary: establishing shot, medium close-up, insert, POV, slow dolly push-in, crane shot, orbital track, and rack focus.

Atlabs makes the verb the main lever: a singer who struts, trudges or stalks across the stage gives the model performance and mood, where "walks" gives it nothing. Dreamina's 2.5 guide uses the same grammar, CU, MCU or WS plus one move such as push-in, dolly left or handheld drift, and says to ask for a match cut or whip pan at a timecode, which is how you land a cut on the downbeat. Our 28-prompt camera movement library maps moves to song sections.

What should you avoid in a Seedance 2.5 prompt?

Avoid adjective stacks, conflicting camera instructions, too many actions for the duration, vague identity rules, and references that disagree with each other. Dreamina groups negative constraints at the end and ties each one to a real production need. InVideo names the 2.0 failure modes as too many style adjectives, unclear reference roles, and low-quality source clips.

Dreamina's troubleshooting fixes, in our words:

• Face or wardrobe drifts: name one primary reference, list the fixed traits, repeat them at each major beat.
• The output looks generic: swap broad words such as cinematic for a concrete location, a practical light source, a lens behavior and a texture.
• Sound feels disconnected: attach each cue to an action or timecode and say whether the music rises, holds or resolves.
• On-screen text wobbles: the model struggles to redraw typography in motion, so render clean footage and add lyrics in the edit.

Dreamina's debugging rule is to change one control layer at a time: references and identity first, then shot structure, then camera, then style and audio. PrompTessor adds one sequence mistake: changing wardrobe or location wording between shots when you did not intend a visual change.

How do you prepare a music video before you render?

Storyboard the song into 30-second blocks, write a canon for the performer, cut the audio into matching slices, and build the reference pack before you open the generator. A 3-minute song is 180 seconds, so six 30-second clips is the floor; at Dreamina's 6 to 8 seconds per beat, that is roughly 22 to 30 beats to plan.

Dreamina's four-beat structure for a 30-second scene fits a verse or chorus: a wide opener to about 6 seconds, medium close-up development to 14, a moving shot or insert to 24, and a close-up callback to 30. It also recommends cheap previs first: neutral materials, a fixed lens, two or three takes where only the camera move changes, and start and end frames so the clip does not wander. OiiOii's founder Naonao put the cost plainly at the August 7, 2026 launch: "a clean 30-second shot takes real prompting craft". That craft is the prep.

1. Timestamp the song's sections and assign each a 30-second block; our storyboard guide has the shot list format.
2. Approve two or three portrait stills of the performer and one location still.
3. Write the canon: fixed traits, wardrobe, location and one lighting rule, pasted into every prompt.
4. Cut an audio slice per block and name the files to match the @Audio handles.
5. Write one 8-part prompt per block, timed beats, one camera move per shot, then render the chorus block first and fix one layer at a time until it holds.

Two honest limits: thirty seconds is one clip, so a full video is still six or more renders plus retakes, and a singing close-up that must match the track across cuts remains the shot most likely to miss. On where to render, Dreamina carries Seedance 2.5, and Seeddance (seeddance.video), an independent site that resells Seedance models, is another route; our /go/seedance links go to seeddance.video and we earn a commission if you subscribe. For what a 30-second block costs there, read our Seedance 2.5 pricing and credits guide.

Next step: mark the first chorus of your song and write one 8-part prompt for it from the table above. Render that block, debug it one layer at a time, then reuse its canon lines in every other block and follow the Seedance music video walkthrough for assembly.

Primary sources

• Dreamina, Seedance 2.5 Prompt Guide: Formula, Examples & Pro Tips (August 18, 2026)
• InVideo, Seedance 2.0 Prompt Guide (With Examples) (June 9, 2026)
• Atlabs, prompting guide for Seedance 1.0 Pro (undated, read October 25, 2026)
• PR Newswire, OiiOii AI launches full-capacity Seedance 2.5 (August 7, 2026)
• PrompTessor, Seedance prompt generator and prompting techniques (last substantive review September 5, 2026)

Estimate your render cost with our free credit calculator.

Frequently asked questions