Seedance 2.5 AI Video Generator

ByteDance’s newest video model, for shots where skin, fabric and light have to hold up in close-up, with sound made in the same pass.

Write your idea first. You sign in when you press Generate.

Short answer: Seedance 2.5 is ByteDance’s newest AI video model, shown on June 23, 2026 and launched on July 31, 2026. It renders finer textures, skin and light than Seedance 2.0, tells a story across several shots in one clip, and makes the soundtrack together with the picture. Here it runs from the box above: describe a scene, start from a photo, hand it reference images or two frames, and pick 5, 10 or 15 seconds at 480p, 720p or 1080p. The MP4 comes back with sound, unwatermarked, after a few minutes of rendering, from 97 credits a clip. It is paid with credits; to try an idea first, the free Basic engine in the same menu gives you one free video and three free images a day.

How to use Seedance 2.5 online

Four steps from idea to MP4. Google sign-in comes up once, the first time you generate.

  1. 1

    Pick how the clip starts

    A sentence or a single photo (Text/Image to Video), up to 9 pictures of a character, product or set (Reference to Video), or a fixed opening and closing image (Frames to Video).

  2. 2

    Describe surfaces, light and sound

    Subject and action first, then what the camera does, the light and the materials, and last what you should hear. Seedance 2.5 has the detail to draw what you name.

  3. 3

    Set length and resolution

    5, 10 or 15 seconds at 480p, 720p or 1080p. A 5 second clip at 1080p is 535 credits, so test the prompt at the lowest size first.

  4. 4

    Generate and download

    Leave the tab if you like: the clip keeps rendering on our servers and lands in Assets as an MP4 when it is done.

Seedance 2.5 key features

Official reference clips from ByteDance. Suggested prompts are separate from the clips. Vendor showcases may use settings and inputs beyond those offered here.

  • Opera costumes, facial detail and moving fabric in a staged performance.
    Try a related prompt

    Extreme close-up of a woman’s face by a rain-streaked window at night, neon reflections moving across her skin, a single tear forms and falls. Soft rain, a distant siren.

    Close-up detail: skin, eyes, fabric, light

    ByteDance says Seedance 2.5 was refined for textures, skin and eyes, lighting and color. Pores, knitwear, wet stone and brushed metal look photographed rather than smoothed, which is why it carries the "Top quality" tag on Genvima.

  • A man follows a boy out of a subway. Finished result from the official video-extension example.
    Try a related prompt

    Shot 1: a baker opens an oven before dawn, warm light spills out. Shot 2: close-up of a crust cracking as she breaks a loaf. Shot 3: the first customer smiles through the shop window. Oven hum, crackling crust, a door bell.

    Stories told across several shots

    The model plans a clip as a sequence: setup, a turn and a payoff, with cleaner cuts and scene changes than 2.0. Number the shots in the prompt and it keeps the same people and place from one to the next.

  • A singer moves from backstage to a live performance with synchronized sound.
    Try a related prompt

    A street musician tunes a cello under a stone bridge, then plays three deep notes. Echo off the arch, footsteps of passers-by, a pigeon taking off.

    Sound made with the picture

    Seedance 2.5 generates audio and video together and in sync. ByteDance also says it cut down on two things that spoiled earlier clips: subtitles nobody asked for and background music laid over everything.

  • A singer, musicians and an audience in a concert scene guided by multiple references.
    Try a related prompt

    A grandmother and her grandson cook dumplings at a kitchen table. She says "Fold it like this." He laughs and shows her a lopsided one. Sizzling pan, a radio playing quietly.

    Several characters that stay themselves

    Faces, clothes and voices of more than one character hold steady in busy scenes, the point where earlier models swapped details between people. Two or three named characters work best.

  • A girl travels through an underwater world and into space. Finished animation from the official motion-reference example.
    Try a related prompt

    Use @image1 as the model and @image2 as the coat. She walks through falling snow on a quiet street, turns and looks over her shoulder. Snow crunching underfoot, a distant tram.

    References for your own cast and products

    Reference to Video takes up to 9 images here: a face, an outfit, a product, a room. Name them in the prompt as @image1, @image2 and so on, and the shot is built around them.

  • Close-ups and overhead shots of an animated breakfast scene. The generated camera-edit result.
    Try a related prompt

    A sunny kitchen breakfast scene. Begin close to an egg sizzling in a pan, track along the counter toward a cup of coffee, then pull back to show the cook. Keep the camera movement smooth and the kitchen layout consistent.

    Camera direction and shot timing

    Describe the order of camera moves and the moment each action should happen. A close-up, a tracking move and a wider reveal can give a short scene a clear sequence. The official example shown here demonstrates camera editing; use its movement as a reference when writing a new generation prompt.

Seedance 2.5 specs and credits on Genvima

ModelBest forResolutionClip lengthSoundModesFrom
Seedance 2.5Top quality480p, 720p, 1080p5, 10, 15 sYesText/Image, Reference, Frames97 credits
Seedance 2.0Best overall480p, 720p, 1080p5, 10, 15 sYesText/Image, Reference, Frames66 credits
Seedance 2.0 FastFaster480p, 720p5, 10, 15 sYesText/Image, Reference, Frames40 credits
Seedance 2.0 MiniBest value480p, 720p5, 10, 15 sYesText/Image, Reference, Frames19 credits
Lengths, sizes and credits as the generator offers them today, quoted for the shortest, smallest clip. Rows for the Seedance 2.0 family show what the extra detail costs.

What you can make with Seedance 2.5

Six jobs where the extra detail earns its price. Copy a prompt, swap in your subject, and render.

  • Hero shots for ads

    The one shot that opens the spot and has to look expensive: a face in golden light, a watch on a wrist, a drink being poured.

    Try this promptSlow motion close-up, sparkling water pours into a glass full of ice on a marble bar, backlit by late sun, bubbles rising. Fizz and ice clinking.

  • Beauty and fashion

    Skin, hair and fabric are the hardest things for a video model and the easiest place to see quality. Try it for lipstick, skincare, coats and hairstyles.

    Try this promptClose-up of a woman applying red lipstick in a mirror, soft window light from the left, the camera eases toward her lips. A quiet room, a cap clicking shut.

  • Food and drink

    Steam, condensation, melting cheese, crisp crusts. Close framing and slow motion show what the model renders.

    Try this promptMacro shot, a fork pulls a slice of pizza away and the cheese stretches in long strands, steam rising, warm restaurant light. Sizzle, a crowd murmuring.

  • Short narrative scenes

    A mini trailer or the opening of a series: three numbered shots with a turn in the middle and sound already in place.

    Try this promptShot 1: a lighthouse keeper climbs a spiral staircase with a lantern. Shot 2: close-up as he sees a ship’s light in the storm. Shot 3: the beam swings across the waves. Wind, rain, a foghorn.

  • Product reveals

    Reference the product, then ask for a reveal with controlled light. Keep labels and shapes described exactly.

    Try this promptUse @image1 as the headphones. They rest on dark velvet, a beam of light slides across them and the camera orbits half a turn. A low synth swell.

  • Final renders of tested prompts

    Find the shot on Seedance 2.0 or Mini, which cost less, then render the version people will see on 2.5. The prompts carry over unchanged.

    Try this promptThe same prompt you tested on Mini, unchanged, at a higher resolution.

Seedance 2.5 vs Seedance 2.0, Fast and Mini

All four take the same prompts. What you pay for on 2.5 is detail; what you save on the others is credits.

Seedance 2.5Seedance 2.0Seedance 2.0 FastSeedance 2.0 Mini
Best forClose-ups and hero shotsFinal clips with people and soundQuick turnaroundsDrafts, volume and trend videos
Resolution480p, 720p or 1080p480p, 720p or 1080p480p or 720p480p or 720p
Clip length5, 10 or 15 s5, 10 or 15 s5, 10 or 15 s5, 10 or 15 s
SoundYes, generated with the videoYes, generated with the videoYes, generated with the videoYes, generated with the video
Ways to startText/Image, Reference, FramesText/Image, Reference, FramesText/Image, Reference, FramesText/Image, Reference, Frames
Credits, shortest clip at the lowest size97664019
Pulled live from the generator, so a price or length change shows up here the same day.

Seedance 2.5 prompt guide

Seedance 2.5 reads the same prompts as 2.0, but rewards a few extra words about materials and light. A reliable order: who, doing what, where, camera, light, sound.

  • Name the surfaces. Wet cobblestones, brushed steel, wool, freckles, condensation. The model has the detail to draw them.
  • Name the light. Window light from the left, golden hour backlight, a single desk lamp, neon from outside.
  • Keep the camera slow when detail matters. A slow push in or half an orbit shows texture; a whip pan hides it.
  • Number the shots for a sequence: "Shot 1: … Shot 2: … Shot 3: …", one line each.
  • Finish with audio: name two or three sounds the scene would really make, or put one short spoken line in quotation marks.
  • Ask for what you do not want only when it appears: "no subtitles", "no music".

Seedance 2.5 image to video tips

Text/Image to Video takes one photo as the first frame. Seedance 2.5 keeps the detail of the picture, so the photo matters as much as the prompt.

  • Use a sharp, well-lit photo at least as large as the resolution you pick. Blur and heavy compression come back as mush in close-up.
  • Match the ratio. A vertical photo for a 9:16 clip, a wide one for 16:9, so nothing important is cropped away.
  • Describe only what changes. The photo already shows the face and the room; the prompt says what moves, what the camera does and what you hear.
  • Keep the first second calm. "She slowly turns her head" works better than a sudden jump, because the model starts from a still picture.
  • Use Frames to Video for a fixed ending. If the clip has to land on a given picture, add it as the last frame instead of describing it.
  • Use Reference to Video when the photo is not a first frame: a face, a product or an outfit you want in a new scene.

Is Seedance 2.5 worth it over 2.0?

Seedance 2.5 starts at 97 credits a clip against 66 for Seedance 2.0 and 19 for Mini. The difference shows in close-ups, on large screens and at 1080p.

For a wide shot that will be watched on a phone, most viewers will not see it, and 2.0 or Mini is the better buy. For the close-up that leads an ad, a beauty shot or a product reveal, use 2.5.

What ByteDance announced

ByteDance first showed Seedance 2.5 at its Volcano Engine conference on June 23, 2026, and its Seed team launched it on July 31, 2026 in Jimeng and Doubao. The launch post describes 30 second clips with audio in one generation, multi-round extension to videos of several minutes, up to 30 images, 10 video clips and 10 audio clips as references in one request, and editing that can target a time range inside a clip.

ByteDance gives no output resolution in the post. Here, Seedance 2.5 renders 5, 10 or 15 second clips at 480p, 720p or 1080p from text, a photo, up to 9 reference images or two frames, and the box above lists every setting a clip can use.

Limits worth knowing

  • Complex motion and crowds. ByteDance says the physics of complex motion and scenes where many subjects interact still need work. Keep it to two or three people.
  • Lettering. Signs and captions inside the frame often come out garbled; put words on in your editor.
  • Length. 15 seconds is the most one clip can be here. Longer videos are several clips joined in an editor.
  • Famous faces. Seedance 2.5 may decline a prompt or reference built around a recognizable real person. When it declines, the credits return to your balance.

Seedance 2.5 vs Kling 4.0 and Veo 3.1

Kling 4.0 is the pick for long, steady camera moves and fast action. Veo 3.1 is very strong at realistic light and dialogue in 8 second clips. Seedance 2.5 wins on close-up texture and on multi-shot stories with sound.

Many people mix them: draft on Seedance 2.0 Mini, render the hero shot on 2.5 and the moving shots on Kling. One credit balance covers all of them (see pricing), so switching models mid-project costs nothing extra.

FAQ

What is Seedance 2.5?

Seedance 2.5 is ByteDance’s newest AI video model and the successor to Seedance 2.0. It was shown on June 23, 2026 and launched on July 31, 2026. It generates video and sound together and is tuned for fine detail and multi-shot storytelling.

When was Seedance 2.5 released?

ByteDance’s Seed team launched it on July 31, 2026, after a first showing at the Volcano Engine conference on June 23, 2026.

How can I use Seedance 2.5 online?

Open the model menu above, choose Seedance 2.5, describe the shot or upload a photo, then press Generate. A Google sign-in and credits are all it takes; ByteDance accounts and API keys stay out of it.

Is Seedance 2.5 free?

No. Seedance 2.5 is paid with credits: you sign in with Google, buy a plan or a credit pack, and every clip shows its price on the Generate button before you spend anything. To try an idea first, the free Basic engine in the same menu gives you one free video and three free images a day, with no credit card.

How long are Seedance 2.5 videos?

ByteDance’s launch post gives 30 seconds per generation. On Genvima you choose 5, 10 or 15 seconds, at 480p, 720p or 1080p.

Does Seedance 2.5 have 4K?

Not as far as ByteDance has said; its launch post gives no 4K figure. On Genvima the sharpest Seedance 2.5 setting is 1080p.

How many references does Seedance 2.5 take?

ByteDance gives up to 30 images, 10 video clips and 10 audio clips in one request. Reference to Video on Genvima takes up to 9 images.

Does Seedance 2.5 generate audio?

Yes. Speech, ambience and sound effects are generated in the same pass as the picture and stay in sync with it.

How much does Seedance 2.5 cost?

From 97 credits for the shortest clip at the lowest resolution; a 5 second clip at 1080p is 535 credits. Credits come with a plan or a credit pack.

Is Seedance 2.5 free?

No. A Seedance 2.5 clip starts at 97 credits, paid from a plan or a credit pack after Google sign-in, and the exact figure is on the Generate button before you commit.

Seedance 2.5 vs Seedance 2.0: which should I use?

2.5 for close-ups and hero shots where detail shows; [Seedance 2.0](/video/seedance-2-0) (from 66 credits) for everything else and for many variations.

What happens if a Seedance 2.5 clip fails?

A failed or refused render is refunded in full to your credit balance, once, without a support ticket.

Keep exploring

Try Seedance 2.5 now

Write the idea, pick the model, and see the price before you generate.

Back to the prompt box