Seedance 2.5
Seedance 2.5 is an AI video generation model from ByteDance, now available on fal. Seedance 2.5 generates native 30-second video at up to 720p from text, a single image, or up to 50 multimodal references.
Your prompt opens in the playground, where you generate. All content on this page was generated with
Dreamina Seedance 2.5 and published by BytePlus. Not generated via fal API.
What can Seedance 2.5 do?
A full 30-second scene in one AI video generation
Dreamina Seedance 2.5 generates up to 30 seconds as one native take, rather than short clips stitched together.
- The model reasons about the whole shot at once, so identity, clothing, prop ownership and spatial direction hold from the first frame to the last instead of drifting at the joins.
- Detailed prompts produce finished footage and assets you can use as they come out.
- For a shot with several beats, BytePlus advises writing it as consecutive stages rather than one long paragraph: give each stage a single primary change, and state what should be visible on screen when it ends.
Bring 50 of your own references into one AI video generation
A single generation can combine up to 50 reference materials, and each one is addressed directly in the prompt.
- Four performers, one per image reference, sing and dance through changing formations in the clip alongside.
- Every one of them lip-syncs to the track supplied as an audio reference, with cuts landing on the beat and VHS grain across the whole piece.
- It opens on an extreme close-up of an eye and closes on an FPV circle around the rooftop, all in one generation.
AI video that moves like filmed footage
Weight, momentum, fabric and hair behave in generated video the way they do in filmed footage.
- Sleeves trace natural arcs through the air instead of sliding through the body.
- A subject placed into a new environment responds to it, from the direction cloth catches the wind to the way light falls across it.
Block the shot in 3D, get a finished AI video render
Lay out a scene in a 3D tool, export the viewport, and let the model finish it as video.
- A coarse blockout of simple geometry locks the action paths, blocking, entrances and exits, camera path, cut points, lighting changes and sound rhythm. Map each geometric object in the prompt to the subject or prop it stands for.
- A fine blockout already carries the full structure, so it is re-rendered with new materials, characters, scenes and style while structure, action, spatial layout and camera treatment hold.
Up to 30 images, 10 videos and 10 audio clips in one AI video
The budget of 50 references breaks down into up to 30 images, up to 10 videos and up to 10 audio clips, with video and audio each capped at 30 seconds in total.
- The model reads composition, scenes, styles, characters and props across them and applies them as instructed.
- It selects the right material per scene rather than forcing all of them into shot at once.
- Multiple characters keep their appearance and their voices, even through group scenes and shot changes.
Green screen AI video editing: change the background, keep the take
Swap what sits behind the subject and tell a different story with the same performance: a training ground becomes a locker room, then a packed stadium.
- The subject reacts to the new surroundings in how clothing catches the air, the way hair sits and the rhythm of a stride.
- The light of the new scene falls across the subject rather than leaving them lit for the old one.
Pace an AI video generation with timestamps
Write the prompt as a timed shot list and the model spends the running time where you tell it to.
- 0 to 2.5s pushes in from inside the drum, 2.5 to 5s lands the first line of dialogue.
- Ranges run consecutively without overlapping.
- A timestamp gives an event its share of the running time rather than marking a frame-accurate cut.
AI video generation in multiple languages
A story is generated natively in its own language rather than dubbed after the fact.
- Major languages are covered, among them Chinese, English, Spanish, Indonesian, Malay, Thai, Arabic, Portuguese, Vietnamese, Japanese and Korean.
- Improved stability keeps stray subtitles and unwanted background music out, leaving a clean take ready for text and music editing.
How to use Seedance 2.5
Three steps from a blank prompt box to a finished 30-second video with sound.
What are the examples of Seedance 2.5?
Films BytePlus published with Dreamina Seedance 2.5. Every one links to the post it came from.
A seamless circus long take
The curtain rises on a Victorian circus. A magician takes the stage, performers fill the ring, and the spectacle builds into carefully choreographed chaos, the sequence unfolding continuously from one act to the next.
Source: @BytePlusGlobal on X.
Classic moments. Now within reach.
BytePlus shared a story with Michael Owen, revisiting defining moments from his career, from signature moves on the pitch to open-ended creative possibilities.
Source: @BytePlusGlobal on X.
The missing sock
A photoreal corporate comedy. A boardroom talks in circles while a man hunts for a lost sock at home, and the camera dives into the washing machine drum after a voice calling from somewhere inside it. Dialogue and subtitles throughout. This is the film behind the timed shot list above.
Source: @BytePlusGlobal on X.
Three minutes, one cast, sixty shots
An otter courier, a stag beetle rolling a crushed school bus, and a tapir in a bubble-canopy plane hold their designs across more than sixty shots and four lighting environments, from a teal glass city to a warm diner to a field of giant poppies. A directed short film with its own credited crew.
Source: @BytePlusGlobal on X.
Who is Seedance 2.5 for?
A full 30-second spot, not a snippet
Thirty seconds in one take is a complete ad rather than a fragment to cut around. Feed in product shots and a script, and generate a version per aspect ratio without booking a shoot.
Direction that survives the whole shot
Block a scene in a 3D tool and let the model finish it, or pace the take with timestamps. Characters, wardrobe and spatial direction hold from the first frame to the last, across cuts.
On-brand content in one pass
Combine reference images, a voice track and a prompt to keep the same character and the same look across posts, in any of eleven-plus languages, natively rather than dubbed.
Start building with the Seedance 2.5 API
Three endpoints, all generating native 30-second video at up to 720p.
Dreamina Seedance 2.5 generates native 30-second single-shot video at up to 720p from a single text prompt, reasoning about the whole shot at once so motion, lighting, and subject identity stay coherent from first frame to last.
Dreamina Seedance 2.5 animates a single still into a native 30-second clip at up to 720p, extending one frame into continuous, coherent motion without the drift or stitching of shorter multi-clip workflows.
Dreamina Seedance 2.5 generates video from up to 50 multimodal references images, video, audio, and style inputs, locking a character, set, and palette across a full 30-second take for production-grade consistency.
Seedance 2.5 FAQ
What is Seedance 2.5?
Dreamina Seedance 2.5 is ByteDance's next-generation video generation model. It generates native 30-second video at up to 720p and combines up to 50 multimodal references in a single generation.
Can I use Seedance 2.5 on fal?
Seedance 2.5 API is now available on fal. There are three endpoints: text to video, image to video and reference to video.
Were the videos on this page made on fal?
No. Every clip was generated with Dreamina Seedance 2.5 and published by BytePlus, and the films link to the post each came from. None were generated via fal API.
How long can a Seedance 2.5 clip be?
Dreamina Seedance 2.5 generates native 30-second video at up to 720p, reasoning about the whole shot at once rather than stitching shorter clips together. The circus long take above is BytePlus's own demonstration of a sequence unfolding continuously from one act to the next.
What resolution does Seedance 2.5 output?
Dreamina Seedance 2.5 on fal generates at 480p or 720p, selected per request, in any of seven aspect ratios from 21:9 to 9:16. Duration is set independently, at any whole second from 4 to 30.
Can I control the first and last frame in Seedance 2.5?
Yes. Image to video takes a starting image, and optionally an end image as well, so you can pin both ends of the shot and let Dreamina Seedance 2.5 generate the motion between them.
Do the clips include audio?
Yes, every clip on this page has a soundtrack. Note that the animated short is a directed production with its own credited sound designer, so its mix is not a demonstration of model-generated audio.
How does Seedance 2.5 relate to Seedance 2.0?
Dreamina Seedance 2.5 is the next generation of the Dreamina Seedance family. Both are on fal — you can also use Dreamina Seedance 2.0 today.
How do I get started?
Open text to video and run a prompt in the playground, or call the endpoint directly from the API docs on that page. For volume pricing or enterprise terms, use the contact form below.
Get in touch about Seedance 2.5
Talking to us about volume pricing, enterprise terms or eligibility for Dreamina Seedance 2.5? Leave your details and our team will get back to you.
Contact Sales
Last updated: . Dreamina Seedance 2.5 model details and availability are accurate as of this date.