Storyboard a 30s Seedance 2.5 pass

Seedance 2.5 can do a 30-second single pass at a 720p ceiling, native audio, from 34 credits (was 67). Seedance 2.0 still tops at 15 seconds, 1080p, from 44 credits. Write the beat list before you write the prompt. One generate can hold one 30-second pass on 2.5. On 2.0, one generate is one beat (or one 15-second stretch), not a 30-second ad squeezed into a 15-second card.

This is a storyboard procedure. It is not a model-card restack, and it is not “prompt longer.”

2.5 versus 2.0 is a duration decision

Pick the row from the length you actually have to hold, then from resolution.

Row	Job	Duration	Resolution	Native audio	Credits
Seedance 2.5	T2V	4–30s	720p ceiling	Yes	From 34 (was 67)
Seedance 2.0	T2V + R2V	4–15s	1080p	Yes	From 44

If the brief is a 30-second UGC or a one-minute Creativity Program piece that you will cut, 2.5 is the door that can attempt the 30-second hold in one job. You give up 1080p. If the brief is a 12-second proof at 1080p, 2.0 is the sharper plate and the wrong row to force a 30-second wish into.

Do not “stretch” 2.0 by writing five scenes into a 15-second prompt. That is how a generate invents jump cuts: a hand teleports, the pack is suddenly open, the CTA lands before the claim. One generate, one beat on 2.0. Two or three 2.0 shots, concatenated, is a storyboard. One 2.5 pass with a beat list is also a storyboard — the storyboard just lives in the prompt instead of in the timeline.

Native audio is on for both. The 30-second pass will talk and make room tone for thirty seconds. If the line must be word-perfect, write the line in the beat list as quoted speech, then listen. A wrong 30-second baked line is expensive to keep. Silent Turbo plus TTS is the other path when the copy is a contract.

The AI video generator will run whichever row you pick. It will not silently upgrade 2.5 to 1080p or 2.0 to 30 seconds. The card is the spec.

Write the beat list before the prompt

A 30-second pass without a beat list is a paragraph. Models compress paragraphs. You get a cousin of the first sentence for half the runtime and a rushed end card.

Write beats on paper (or in the brief) with clocks that add to 30, not 45.

Example for a product UGC:

0–3s — Hook. Face or pack. One claim. No setup.
3–10s — Problem. The specific fail from last Tuesday, not “life is busy.”
10–20s — Proof. Hands on the SKU. The verb: pour, click, wipe, apply. Label readable.
20–26s — Result. After-state. Same location if you can. No new character.
26–30s — CTA. Pack + next step. Spoken line matches on-screen line you will burn later.

That list is the prompt’s spine. It is also the QC spine. When the file comes back, you score beats, not vibes.

Rules for the list:

One verb per beat. “Walks in, pours, winks, holds the pack, and points at the URL” is five verbs. It will jump.
Name the SKU once, then keep it. If you need the bottle identical across beats inside one T2V pass, you are asking a lot of a text-to-video row. A locked still plus I2V/R2V is the identity path. 2.5 T2V is the duration path. If both are required, still + a reference-capable row, or still + several I2V shots.
Quoted dialogue per beat, if anyone speaks. Native audio will invent speech if you do not pin it. Pin it.
Camera per beat: close-up, medium, pack insert. Do not write “cinematic masterpiece.” Write “handheld medium, then insert of the label.”
What must not happen: extra people, extra logos, extra rooms. Negatives belong in the list because a 30-second model has time to wander.

If you cannot fill 30 seconds without repeating beat 3 twice, you do not have a 30-second idea. Cut to 15 and use 2.0 at 1080p, or write a real second proof.

Seedance 2.5 was announced 23 June 2026 (Volcano Engine FORCE) with a native ~30s single pass. Hosts disagree on 4K; this catalog’s ceiling is 720p. Do not write 4K into the beat list as if the card listed it.

One 30s pass versus several shorter shots

Two production shapes. Pick one on purpose.

Shape A — one 2.5 generate. You need a continuous 30-second feel: one room, one person, one conversation with the camera. The beat list goes into one prompt, in order, with timestamps or “then” marks the model can follow. You accept 720p. You accept that a miss in beat 4 means regenerating the whole 30 seconds (557 credits again at 720p), not recutting beat 4.

Shape B — several shots. You need a cut. Hook on a face, proof on hands, CTA on a pack. Each shot is one generate. 2.0 at 1080p for the hero proof (from 44 credits, up to 15s). Cheap silent Turbo (from 22 credits) for throwaway hook tests. Concatenate. Caption once.

Shape A is cheaper in concatenating and worse in isolation. Shape B is more edits and lets you throw away a 4-second miss without binning the proof.

Do not mix them accidentally: a 30-second 2.5 pass that you then cut into six pieces because beats 2 and 5 failed. You paid for continuity you threw away. If you already know you will cut, start in Shape B.

Procedure for Shape A:

Lock the beat list and the quoted lines.
Pick 9:16 if the destination is TikTok / Reels / Shorts. Do not generate 16:9 “for flexibility.”
Run Seedance 2.5 at 30s and 720p, 557 credits.
Watch once for story. Watch again for identity: does the pack stay the pack? Does the face stay the face?
Scrub every beat boundary. A teleport at 0:10 is a failed pass, not a style.
If beat 4 is the only miss and the rest is gold, you still regenerate — or you cut to Shape B for beat 4 only and live with the cut. Do not “cover” the teleport with a caption.

Procedure for Shape B when 2.0 is the cap:

Same beat list.
One generate per beat. 2.0 if you need native audio and 1080p. Duration per shot ≤15s.
Match wardrobe and location in the prompts; better, lock a still and I2V/R2V the proof shots.
Concatenate on the timeline. Trim dead air.
One caption pass on the combined file.

QC the 30s take before captions

A 30-second native-audio file is a lot of speech to caption wrongly. QC picture and sound first.

Picture:

Label readable at 720p on a phone. 2.5’s ceiling is 720p; if the type on the pack is mush, the pass failed even if the motion is pretty.
Identity. Mid-clip morphs are common when one prompt asked for too many verbs.
Hands. Count fingers on the pour and the hold.
Jump cuts you did not author. Same shot, time missing, subject did not travel through it.

Sound:

Quoted lines match the beat list. Extra clauses the model ad-libbed: keep only if legal and on-brand.
Room tone consistent. A new music bed at 0:18 is a miss if you did not ask for one.
Loudness. A whispered CTA after a shouted hook is a mix problem you will feel on captions too.

Only then transcribe and burn. Captions on a 30-second miss are a second bill on a file you will bin.

If Shape A cannot hold the SKU, stop using T2V for the proof beats. Lock a product still and run I2V or R2V for those seconds. 2.5’s gift is duration, not pack fidelity.

If you need the studio to run the 2.5 pass from a beat list, sign up. Bring the clocks and the quoted lines. Do not bring “make a 30s viral ad.”

FAQ

Can Seedance 2.0 do a 30-second single pass?

No. On this catalog 2.0 lists 4–15 seconds at 1080p, from 44 credits. One generate is one beat (or one ≤15s stretch). Storyboard multiple shots, or use 2.5 for a 30s 720p pass.

Why not always pick 2.5 if it is longer and starts cheaper in credits?

Because the ceiling is 720p and a 30-second miss costs the whole pass, 557 credits at 720p. A 12-second 1080p 2.0 proof is the better object when you do not need thirty seconds.

Should I put captions in the 2.5 prompt?

No. Native audio is the soundtrack. Captions are a burn-in after you approve the 30-second file. Prompted type is texture you cannot restyle.