Seedance 2.5 is a 30-second pass

Seedance 2.5 is a thirty-second single pass, not a stitcher and not an Omni extend. On Versely the published card is Seedance 2.5: text-to-video, every whole second from 4s through 30s, 720p ceiling, 47 credits, native audio on. ByteDance announced the line at Volcano Engine FORCE on 23 June 2026. The catalog description is the job: one text prompt, one continuous shot, motion and identity held from first frame to last because the model reasons about the whole take at once. If you write five ads in one prompt, you will get a thirty-second argument. If you write one verb a camera could hold, you will get a pass.

This is a creator routing guide for that card. It is not an OEM product-film recipe, not a 4K promise (this row’s ceiling is 720p), and not Omni 1.1’s forty-second extend.

What the card actually sells

Read the fields before you brief a “long AI video.”

Category: text-to-video. Not reference-to-video on this published row. If your brief is a stack of stills, you are on Seedance 2.0 (4–15s, 1080p, 69 credits, T2V + R2V, native audio) or Veo R2V, not 2.5.
Duration menu: 4s through 30s, every second. Thirty is the ceiling, not a default. A 12-second action should be 12s. Padding to 30s is how identity thins in the last third.
Resolution: 720p. Hosts elsewhere have claimed 4K for 2.5. This card does not. If the buyer wrote 4K in the insertion order, this is the wrong row. Veo 3.1 and Kling Video V3 4K Text to Video are 4K doors with their own length caps.
Credits: 47. That is the catalog figure on the model page. Credits, not USD.
Audio: true. Native audio means the track is born with the picture. A silent prompt is not a silent file. Write the room: hiss, clink, one line, or “no speech, no music.”

The point of a single pass is coherence. Extend chains (Omni 1.1: 10s of prior context, 10s steps, 40s cap on Google’s surface; Veo extend: a Veo continuation) hand off at a seam. 2.5 does not hand off. Lighting, walk cycle, and product identity have one generate to survive. That is why you brief one continuous action, not a sequence of scenes separated by commas.

Seedance 2.0 remains the 4–15s, 1080p cinematic cousin. Use 2.0 when fifteen seconds at 1080p is enough and you might need references. Use 2.5 when the action is longer than 2.0’s cap and 720p is acceptable. Do not run both “to see.” The length is the pick.

Write thirty seconds as a camera hold

The brief that works is a shot a camera operator could execute without a cut.

Lock five things, then stop:

Who stays in frame. One subject, wardrobe already decided. A recast at 0:12 is a new ad.
The verb. Walks the aisle. Turns the pack. Pours. Tamp, lock, extract, plate. Not “then a testimonial, then a logo, then a CTA.”
The camera. One move that can complete in the duration you picked: slow dolly, locked wide, a follow that never cuts.
The light. One setup. Window left. Practicals. Do not relight at the midpoint.
The sound. One acoustic space. If someone speaks, they speak as part of this action.

A prompt that behaves:

A barista in a navy apron stands at a stainless counter in late-afternoon window light. She tamps an espresso puck, locks the portafilter, starts the shot, watches the crema, and sets the cup on a saucer toward camera. Slow push from a medium wide to a tight on the cup. Soft room tone, the machine hiss, one ceramic clink. Continuous take, no cuts, 22 seconds.

Then pick 22s on the menu. Do not pick 30s “for safety.” The extra eight seconds are empty time the model will fill with fidget.

A prompt that fails:

Open on the logo, cut to the founder talking, then the product on a turntable, then a lifestyle kitchen, then the CTA with the URL.

That is an edit. 2.5 will try to be all five and land on none. Make five clips on shorter rows, or write one of those five as the thirty-second pass and shoot the rest separately.

Duration pick, practical:

4–8s. You probably wanted Veo, Omni Video, or Kling Turbo. 2.5 can do it; 47 credits is a lot for a hook you could have bought cheaper.
9–15s. Overlap with Seedance 2.0. Prefer 2.0 if you need 1080p or references. Prefer 2.5 if you already know you will grow the same action past 15s and want one model for the series.
16–30s. This is why the row exists. One hold. One verb.

Open it in the AI video generator. Confirm 720p and 47 credits on the card before you write “4K 30s” in a Slack thread.

What 2.5 is not

Not Omni 1.1 extend. Google’s 27 August 2026 Omni 1.1 Flash extend uses up to ten seconds of prior context and continues in ten-second increments to forty seconds cumulative, on Google’s surface, model id gemini-omni-1.1-flash. That is a sequel writer. 2.5 is a first pass that can already be thirty seconds. If you do not have an Omni clip to continue, you do not need Omni extend. If you need one file that was never a chain, 2.5.

Not Veo extend. VEO 3.1 Extend Video continues a Veo take. Veo generate itself is 4 / 6 / 8 seconds. A “30-second Veo film” is still a stitch or an extend chain. Call it that.

Not a talking-head spec just because audio is on. Native audio here is room, foley, and speech the model invents as part of the take. If the job is a specific line that must land for a media buyer, Veo 3.1 is still the dialogue-shaped cinematic row (4 / 6 / 8s, 4K tier, 40 credits). 2.5 can speak; Veo is the contract when the mouth is the product.

Not 4K on this catalog. 720p. Upscalers exist elsewhere in the roster. Do not promise a native 4K 2.5 export from a 720p ceiling.

Not reference-to-video on this published card. Categories: text-to-video. If the SKU must match three photos, Veo R2V or Seedance 2.0 R2V.

Not captions. Thirty seconds of speech still needs burned-in captions for silent autoplay. 2.5 will not typeset safe area.

Not an Article 50 exemption. A realistic synthetic of a real person still needs a label when you deploy it to an EU audience. Article 50 disclosure.

When to spend 47 credits, and when to refuse

Spend them when the length is the content: a process that is dishonest at eight seconds (brew, assemble, walk-through, pack-out), a talking demo that is one action, a product that has to live on screen long enough to be believed. UGC-shaped holds often want this. The UGC glossary is the term; this row is one way to generate a long uncut take that still reads as one room.

Refuse them when:

The idea is a three-second hook. Use a cheap short row: Kling 3 Turbo Text to Video is silent 3–15s, 1080p, 12 credits. Hailuo 2.3 Fast is 6s / 10s image-to-video at 28 credits.
The idea is 4K cinematic dialogue. Veo 3.1.
The idea is four scenes. That is an editor. Generate four short clips.
You have not locked the still or the wardrobe. 47 credits is a bad sketchbook. Lock stills on an image row, then write the 2.5 pass.

After the take, finish like any other clip: crop, captions, disclosure. Do not ask 2.5 to also be the end card. End cards are type. Type is a still or an editor overlay.

If the 30-second pass is the job, sign up and open Seedance 2.5. Pick the duration the action needs. Stop at 720p unless a different row says 4K.

FAQ

Is Seedance 2.5 4K on Versely?

No. The published card lists 720p and 47 credits, durations 4s–30s, native audio, text-to-video. Other hosts have claimed 4K for 2.5. Do not import that onto this SKU.

How is this different from Omni 1.1’s 40-second extend?

Omni 1.1 extend continues an existing Omni clip in 10s steps with up to 10s of prior context, to 40s, on Google’s 1.1 surface (27 August 2026). Seedance 2.5 generates a native 4–30s take in one pass on this catalog. No prior clip required. No 1.1 slug required.

Can I paste a storyboard of five shots into one 2.5 prompt?

You can. You should not. The model is a single-pass hold. Five shots is an edit. Write one continuous action, or run five shorter generates and cut.