Seedance 2.0
Live now · Public since July 31, 2026

Seedance 2.5 AI Video Generator

One continuous 30-second take — dialogue, footsteps, and score generated in the same pass as the picture. ByteDance's flagship model is live in this AI video generator: turn text or an image into video with stronger multimodal references (up to 50 inputs), multi-round extensions for longer stories, and lip-synced speech you can direct with plain language.

Live now·Text-to-video & image-to-video·30s single shot·Up to 50 refs·Native audio + lip-sync·480p/720p
Generate With Seedance 2.5See Pricing

Showcase — real 2.5 video, unedited, sound on

Open the AI video generator →
From the community

First tests in the wild

Creators' hands-on 30-second takes, unfiltered — embedded from the original posts, videos play from X.

Posts belong to their authors and are shown via X's embed syndication.

Now Live

Direct It Like You're on Set

What this AI video model actually ships — longer takes, joint audio-video, and reference control built for finished stories, not short demos.

Roll for 30 Seconds Straight

One prompt, one continuous half-minute shot — setup, action, and payoff without stitching four clips. Need more runway? Extend the same take across multiple rounds so multi-minute stories keep a consistent audiovisual language.

Audio Generated With the Pixels

Dialogue with matched lip-sync, footsteps, engines, ambience, and score — all generated in the same pass as the picture, in multiple languages. Wrap spoken lines in double quotes. Audio never adds a credit surcharge.

Up to 50 Multimodal References

Feed the model up to 30 images, 10 video clips, and 10 audio tracks in one generation — faces, wardrobe, products, motion, and even a music bed that can drive pacing. 2.5 reads framing and cinematic intent far more accurately, so characters and looks hold across the full take. This generator sends up to 30 reference images per image-to-video job.

UP TO 50FacesProductsWardrobeLocationsStyle framesMotion refsAudio cues

Model budget: 30 images · 10 video · 10 audio. This generator accepts up to 30 reference images per video.

Start an image-to-video generation →

It Understands the Set

Physics, lighting, and emotional intent from plain language — write "slow dolly-in as she realizes" and get the move, the pacing, and the performance. Timestamp-level and region-level edits let you refine one beat without regenerating the whole clip.

Slow dolly-inOrbit the productGolden-hour glowHandheld urgency

Write the move — get pacing, performance, and fine edits without regenerating the whole take.

Built for

Made for the Work You Ship

Real 2.5 video across every format you ship — product ads, short drama, lifestyle, and social — each generated as one continuous 30-second take with sound.

Product Ads in One Pass

Opening, product moment, end card — a full ad beat in a single 30-second generation instead of three stitched clips. Sound effects land on the action automatically.

Make a Product Ad →

Short Dramas & Cinematic Scenes

Setup, action, payoff — a complete story beat in one unbroken shot. The model holds characters, lighting, and mood steady across the full half-minute.

Shoot a Scene →

Lifestyle & Brand Films

Golden-hour kitchens, morning routines, travel beats — long enough to breathe, with ambient sound generated alongside the visuals.

Create a Brand Film →
21:9

Social Hooks & Vertical Video

9:16, 1:1, 21:9 — six aspect ratios out of the box. Hook, build, payoff in one take, sized for TikTok, Reels, and Shorts.

Make a Vertical Clip →
9:16
NEW

Six Reasons Your Film Comes Together Easier

What Seedance 2.5 changes about day-to-day AI video work.

Open the AI Video Generator

30-Second Single Shot

A full beat with setup, action, and payoff in one continuous take — then extend for multi-minute stories without losing the look.

Native Dialogue & Sound

Lip-synced speech, ambience, and effects generated with the video. No silent-clip workflow, no audio surcharge.

Nothing Drifts

Characters, lighting, and scene style stay consistent across the whole half-minute — the hardest problem in long AI video.

Up to 50 References

30 images + 10 video clips + 10 audio tracks in one generation. Lock faces, products, motion, and music so the look survives the take.

Six Aspect Ratios

16:9, 9:16, 1:1, 4:3, 3:4, and 21:9 — from cinema widescreen to vertical social, no cropping.

Transparent Credit Pricing

From 280 credits for a 5-second 480p clip, scaling linearly to 30 seconds. What you see is what you spend.

30s
Native single-shot length
50
Multimodal references (30+10+10)
∞
Multi-round extensions for longer stories
0
Extra credits for native audio

Built for 30-second stories with sound

Ads, short films, social, e-commerce — every video ships with dialogue, lip-sync, and score generated in the same pass. Storyboard the full half-minute; extend when you need more runway.

Try a Text-to-Video Prompt
CinematicPrompt recipe

Single continuous 30-second take, anamorphic 24fps: a courier weaves her motorbike through a neon-soaked night market — (0–10s) low tracking shot at wheel level, rain kicking off the tires; (10–20s) she cuts down an alley, camera cranes up over the stalls as vendors turn; (20–30s) slow dolly-in as she stops at the harbor edge and lifts her visor. Rain hiss, engine note, crowd murmur fading to one ferry horn; low string pulse building the whole way.

ProductPrompt recipe

One 30-second product film in a single shot: a matte-black espresso machine on slate — (0–10s) macro orbit through drifting steam as the first drops fall; (10–20s) pull back to a slow dolly around the kitchen as morning light sweeps the counter; (20–30s) push-in to the cup, crema swirling, logo catching the light. Beans cracking, steam wand hiss, one soft piano motif landing on the final frame.

LifestylePrompt recipe

Continuous 30-second golden-hour take: a surfer walks out of the water along an empty beach — (0–10s) handheld follow from behind, board under her arm, spray catching the sun; (10–20s) she passes a beach fire where friends look up, camera orbits to their faces; (20–30s) crane up to the cliff line as she drops the board and sits, waves rolling out to the horizon. Gulls, fire crackle, one laugh, warm acoustic guitar swelling to the end.

Drag to browse — steal these structures for your first 30-second take.

How It Works

Generate With Seedance 2.5

Three steps from idea to a finished, sound-on clip.

Describe your shot

For text-to-video, write the scene, action, mood, and camera move, and put dialogue in double quotes for lip-sync. For image-to-video, upload reference images to lock the look, character, or product.

Set the frame

Pick Seedance 2.5 in the model menu, then choose resolution (480p or 720p), duration up to 30 seconds, aspect ratio, and keep audio on.

Generate and export

Roll, review, and download the finished video. Storyboard the full 30 seconds (setup → action → payoff). Draft cheap on Mini first if you're iterating prompts.

Make Your First Video
COMMUNITY

What Creators Ship With Seedance 2.5

From solo YouTubers to marketing teams — workflows built around 30-second takes and native sound.

Lip-synced dialogue in the same pass as the picture is the unlock. I storyboard a full 30-second beat, wrap the lines in quotes, and skip hours of audio post.
Alex RiveraYouTube Creator, 500K subscribers
One 30-second take replaced three stitched clips for our product ads. Characters and lighting hold to the end card — our agency bills dropped overnight.
Sarah ChenMarketing Director, TechFlow
I lock the cast with reference stills and let 2.5 hold them across the whole take. Mood boards used to be six short clips — now they're one continuous scene with sound.
Marcus WebbIndependent Filmmaker
Hook, build, payoff in one 9:16 generation. This is how I ship 20+ sound-on video ads a week without a production crew.
Priya PatelSocial Media Manager
FAQ

Seedance 2.5 FAQ

Straight answers on what 2.5 ships, how it prices, and when to still use Seedance 2.0.

What is Seedance 2.5?
Seedance 2.5 is ByteDance's flagship AI video model. It generates one continuous shot of up to 30 seconds with synchronized audio (including dialogue and lip-sync) in the same pass as the picture, accepts up to 50 multimodal references, and supports multi-round extensions for longer stories. Public release: July 31, 2026 — live on this generator.
What's new compared to Seedance 2.0?
Headline upgrades on 2.5: native 30-second single-shot video (vs 15 seconds on 2.0), multi-round extensions, dialogue with matched lip-sync, and much stronger multimodal referencing — up to 50 inputs (30 images, 10 videos, 10 audio) versus a smaller set on 2.0. Timestamp-level and region-level editing, green-screen workflows, and motion references are part of the 2.5 family rollout. On this site, native 4K remains a Seedance 2.0 path — 2.5 ships at 480p/720p at launch.
How long and what resolution can videos be?
On this generator, 2.5 renders 5, 10, 15, or 30-second videos at 480p or 720p, in six aspect ratios from 21:9 widescreen to 9:16 vertical. Multi-round extension can continue a story beyond 30 seconds with consistent style. ByteDance has also shown longer ultra-long modes on its own apps — we'll expose them when they reach the public API. Need 4K today? Use Seedance 2.0 (2160p) in the same generator.
Does it generate sound and dialogue?
Yes. Every video generates its audio in the same pass as the visuals: dialogue with lip-sync, footsteps, ambience, and score, matched to on-screen action. Wrap spoken lines in double quotes in your prompt. Audio is included at no extra credit cost.
How many references can I use?
The 2.5 model accepts up to 50 multimodal references — 30 images, 10 video clips, and 10 audio tracks. This AI video generator sends up to 30 reference images per job: upload one image to set a starting frame, or two or more to blend faces, wardrobe, and products into a reference-to-video generation (that mode needs a text prompt as well). Video and audio references are not exposed here. Even a single reference image is read more accurately on 2.5 than on 2.0.
Can it keep the same character across a 30-second take?
Yes — that is a core 2.5 design goal. Character, lighting, and scene style are meant to hold across the full take instead of drifting after a few seconds. For strongest consistency, anchor with a reference image, describe the character precisely, and check the last ten seconds before shipping client work.
How much does Seedance 2.5 cost?
Same credit system as the rest of this site: 5s at 480p is 280 credits, 720p is 560, scaling linearly to 30 seconds (e.g. 30s 480p = 1680 credits). Audio adds nothing. Draft prompts on Seedance 2.0 Mini from 40 credits, then re-run winners on 2.5.
How do I get the best results?
Write prompts like a director: subject, action, mood, and one camera move per beat. Storyboard the full 30 seconds (setup → action → payoff) instead of padding a 5-second idea. Put dialogue in double quotes. Use a reference image to lock faces or products. Extend the take when you need a longer scene rather than stitching separate clips.
Is this text-to-video or image-to-video?
Both, in the same AI video generator. Text-to-video turns a written prompt straight into video and is the fastest way to test an idea. Image-to-video starts from reference images and is how you keep a real product, face, or location consistent through the shot. Either mode returns one continuous take with synchronized sound.
Do I need to install anything to use Seedance 2.5 online?
No. The AI video generator runs in your browser — sign in, pick the model, write a prompt, and generate. There is no software to install and no local GPU required. Finished videos are standard MP4 files you download straight from the result page.

Roll Your First 30-Second Take

Seedance 2.5 is live — 30s single shot, native audio with lip-sync, up to 50 references, from 280 credits.

Generate With 2.5 NowSee Pricing