Skip to content
New · Now on Pixo

Seedance 2.5:
AI Video, Directed Down to the Timestamp

Runs of up to 30 seconds — twice Seedance 2.0. Direct every beat with timestamp-level control, then fix any moment with an edit instruction. The Seedance line's biggest jump yet — live now in Pixo's Playground.

Prompt & ReferencesOfficial manual · ultra-wide one-take · text-to-video
Text-only prompt · timestamp script30s · 16:9 · native audio

What's New in 2.5

Nine upgrades that change how you direct

01

Twice the length

Single runs of up to 30 seconds — double Seedance 2.0's 15 — generated in one take, no stitching.

02

Timestamp-level directing

Write instructions against the clock — what happens at 0:03, what changes at 0:12. The model follows your timeline, not its own — down to single frames when you need it.

03

Edit what you generated

Smart edit changes a generated video by prompt, with time-range targeting. Advanced edit adds box-select and brush tools — strip the BGM while keeping voices and captions, remove objects, or shift the camera viewpoint with the model extending space it never shot.

04

A bigger reference budget

Up to 30 images, 10 video clips (30s total) and 10 audio tracks in one prompt — including pure audio-driven generation. And references carry more than motion: emotion, camera language, the whole creative idea.

05

Faces that stay themselves

The "twin face" problem in multi-person scenes is fixed — and so is face-swap drift. Up to 8 characters in frame each keep their own distinct face across shots.

06

Cleaner tracks, wider languages

No more random captions or unwanted BGM on outputs. Language coverage extends across Chinese, English, Spanish, Indonesian, Malay, Thai, Arabic, Portuguese, Vietnamese, Japanese and Korean.

07

Storyboard in, film out

Sketch a multi-panel storyboard — simple line art or stick figures is enough — and the model turns it into a coherent film, with composition, framing and camera moves steered by your prompt.

08

Two clips in, one cut

Hand it two clips and it generates the bridge between them — a seamless transition in whatever style you name.

09

Previz-grade control

White-model (blockout) input directs camera choreography, staging and movement paths with production precision — and green-screen editing turns any background into a clean key for post.

2.5 vs 2.0

What actually changed

FeatureSeedance 2.5Seedance 2.0
Max duration per runUp to 30s per run15s
Resolution720p on Pixo (480p / 720p on Dreamina)Up to 2K
Timestamp-level controlYes — by the second, down to single framesNo
Post-generation editingYes — prompt edits with time-range + box-select/brush toolsNo
Reference inputs30 images + 10 videos + 10 audio; audio-driven mode12 files
Multi-person facesTwin-face fix — up to 8 distinct charactersOccasional face blending
Output hygieneNo random captions / BGMOccasional artifacts
Native audioYesYes
On Pixo todayYes — Playground: generation up to 30s at 720p (editing, extension & first/last-frame not yet)Yes — Playground & storyboard

Specs

Seedance 2.5 at a glance

Output length
Up to 30 seconds per run
Resolution
480p / 720p on Dreamina — 720p on Pixo's Playground
Reference images
Up to 30 per prompt — aspect ratios 0.4–2.5, 300–6,000 px
Reference videos
Up to 10 clips (30s total) — MP4 / MOV, 480p–4K, 24–60 fps, 200 MB each
Reference audio
Up to 10 tracks (WAV / MP3) — audio-only input supported: a track alone can drive the video
First / last frame
First-frame and first+last-frame modes supported alongside omni reference
Timestamp control
Direct by the clock — down to single frames (the official guide scripts a 24-second film across 720 frames)
Native audio
Dialogue, ambience and effects generated with the picture — no stray captions or unwanted BGM
Languages
Chinese, English, Spanish, Indonesian, Malay, Thai, Arabic, Portuguese, Vietnamese, Japanese and Korean

Sweet spots from the official guide: 1–8 subject images, 5–10 second subject clips, and source videos under 20 seconds for edits. Higher counts still work — just expect more retries.

The Cost Question

Longer takes, pricier re-rolls — direct it shot by shot

One take, one spend

A 30-second run costs several times a short clip — one bad take is an expensive bad take.

Shot by shot

Pixo's storyboard breaks your video into shots — generate, review and re-roll only the shots that need it, with any model on each one. Long-form ambition, short-form spend.

How to Prompt 2.5

Direct it like a timeline

  • Write against the clock. "0:00 wide shot of the harbor — 0:08 cut to her face — 0:15 the phone rings." Timestamps are instructions, not decoration.
  • One scene per beat. Longer runs reward structure: break 30 seconds into 4–6 beats rather than one paragraph of vibes.
  • Dialogue must be written out. If a character speaks, give the exact line — audio generates with the picture.
  • Edit with a time range. When fixing a generated video, say where: "from 0:12 to 0:18, change the jacket to red." The rest stays.
  • Name your transitions. The model answers to a full cut vocabulary — hard cut, dissolve, fade, white/black flash, wipe, mask cut, match cut, action cut, relay cut, push/pull, ink-wash. Say which, and where.
  • Cast with a character sheet. Lock people dimension by dimension — age & look, skin texture, facial details, gaze, hair, wardrobe, build & temperament. The ballroom prompt below runs the full formula.
Prompt #1 — Timestamp-Directed One-Take (Glacier to Deep Sea)
Prompt #2 — Multi-Shot Period Drama (Regency Ballroom)
Prompt #3 — Edit Instruction (Moonlight Re-Light)

Read the full prompt guide: character formula, timestamp scripting & 12 transitions →

Why Seedance on Pixo

The strengths that carry over

Prompt-to-Production in Seconds

Describe what you want to see, and the model does the directing for you.

Every Detail Stays Locked

Consistency across composition and visual elements, shot after shot.

Seamless Multi-Angle Narratives

Automatic shot connections with audio-visual sync.

Action That Feels Real

Physics the camera believes — weight, momentum, contact.

How It Works

3 steps to your first Seedance 2.5 video

STEP 01

Go to the Playground

Open Pixo's Playground and start a new generation.

STEP 02

Select Video → Seedance 2.5

Pick Seedance 2.5 as your model, then set aspect ratio, resolution and duration.

STEP 03

Generate & Export

Review the result, re-roll what needs it, and export watermark-free.

FAQ

Questions, answered

What is Seedance 2.5?

Seedance 2.5 is ByteDance's newest video model in the Seedance line, released July 31, 2026. It generates up to 30-second videos in a single run, follows timestamp-level directions, and can edit generated footage by instruction — with up to 30 image, 10 video and 10 audio references per prompt.

What's the difference between Seedance 2.5 and 2.0?

Length and control. 2.0 tops out at 15 seconds; 2.5 doubles that to 30 with timestamp-level directing. 2.5 also adds instruction-based editing, a much larger reference budget, a fix for blended faces in multi-person scenes, and cleaner audio tracks.

How long can Seedance 2.5 videos be?

Up to 30 seconds in a single run — double Seedance 2.0's 15-second cap, and exactly what Pixo's Playground supports today.

Can Seedance 2.5 edit videos?

Yes, two ways: smart edit changes a generated video from a prompt, with time-range targeting; advanced edit adds box-select and brush tools — green screen, BGM separation, object removal and viewpoint changes. These editing tools aren't available in Pixo's Playground yet.

How many reference files can Seedance 2.5 use?

Up to 30 images, 10 video clips and 10 audio tracks in one prompt — and audio-only input is supported, so a track alone can drive the video. Seedance 2.0 topped out at 12 files total.

What reference settings work best in Seedance 2.5?

The official guide's sweet spots: 1–8 subject images (above 5 subjects, single-view images stay most stable — for multiple angles, upload separate shots rather than one collage), subject clips of 5–10 seconds each, and for edits a source video under 20 seconds with 1–5 reference images. You can push past these ranges, but stability drops and you'll retry more.

How precise is the timestamp control?

Down to the frame. Prompts can address moments by the second ("at 0:12, cut to her face") or by frame ranges — the official guide scripts a 24-second sequence across 720 frames. The model holds every mark you set.

How is Seedance 2.5 different from MiniMax H3?

Two different trades. Seedance 2.5 leads on raw capability across the board — timestamp-level directing, instruction editing, and the biggest reference budget of any current model (30 images, 10 videos, 10 audio) — but its runs cost noticeably more. MiniMax H3 stays close on quality at a much lower price per run, and it's open source — with its own edge in voice cloning and TTS across 11 languages. Both are live on Pixo — try them side by side and pick per project.

How do I use Seedance 2.5 on Pixo?

Sign up at pixo.video, open the Playground and pick Seedance 2.5 as your model. New users get 200 free credits on sign-up, and plans are currently up to 55% off.

Direct your first video down to the timestamp

200 free credits on sign-up. No credit card. Full storyboard access.

Try Now

All models. One storyboard. Export watermark-free.

Also on Pixo: Seedance 2.0 — Playground & storyboard →Compare with MiniMax H3 →All AI models on Pixo →