Seedance 2.5: ByteDance's AI Video Model for Social Media
What Seedance 2.5 changes for your Reels and TikToks, when to pick it over Gemini Omni 1.2 or Kling 4.0, and how to choose it scene by scene in Oposto.
By Théo, Oposto team · · Tools
Seedance 2.5 is ByteDance's video model that creates up to 30 seconds in one go, with sound, while following up to 50 image, video and audio references. For social media, it is the right pick when a scene has to look great and stay faithful to a real person or a real product; for a short, simple scene, Gemini Omni does the job for less.
Updated November 9, 2026. In Oposto, you choose Seedance 2.5 scene by scene in a video's storyboard, on the Pro plan and up. For each scene you see what it needs, which models fit and what each quality level costs, before anything is generated. The rest of this article covers what the model can do, based on ByteDance's own pages, and when to pick it over Gemini Omni 1.2 or Kling 4.0.
What is Seedance 2.5?
Seedance is ByteDance's family of video models, from the company behind TikTok. ByteDance's research team presents version 2.5 as a model that generates picture and sound together. You describe a scene in text, add references (photos, videos, audio) and get back a video clip with its soundtrack.
ByteDance offers it through a chat app, a playground and an API on its cloud service, BytePlus ModelArk. You don't need any of these to use it in your videos: a tool like Oposto calls it for you, one scene at a time.
What does Seedance 2.5 change for social media videos?
Four things matter for a Reel, a TikTok or a Short.
Seedance 2.5 in three numbers
- 30 s in one continuous generation
- 50 references at most: images, video and audio
- 2 extensions possible on a clip
30-second shots in one take
ByteDance announces up to 30 seconds in a single generation, with the option to extend the clip twice. Its product page gives a range of 4 to 30 seconds. On social media, that means a whole scene can live in one continuous shot: a product demo, a walk through a store, a small story with a beginning and an end. Fewer cuts means fewer moments where the character or the set changes from one shot to the next.
Up to 50 references
ByteDance's product page talks about 50 reference inputs mixing images, video and audio, so the model can follow a character, product details, scene structure, camera moves and brand assets. The research page adds that a reference video is not only about copying motion: the model also picks up its framing and camera language. That helps you keep the same person across scenes, or borrow the pacing of a video you like.
Sound generated with the picture
Picture and sound come out of the same pass. ByteDance describes audio synced to the millisecond, from ambient effects to speech. For a short video, it means no silent clip to dress up afterwards: footsteps, a door, a mug set down on a table arrive with the image.
Editing without starting over
ByteDance also highlights more reliable editing: swapping a product, a style, a face, the lighting or the background while keeping the motion and composition, plus green-screen editing. On social media, that is what lets you fix one detail of a good video instead of regenerating everything.
Does Seedance 2.5 accept real faces as a reference?
The ByteDance pages reviewed for this article don't publish a rule about photos of real people. In Oposto, a Seedance 2.5 scene accepts a photo of a real person as a reference; it is one of the cases where Oposto recommends it, when someone has to be reproduced from their photo. Earlier versions, Seedance 2.0 and 2.0 Fast, refuse a photo of a real face, and Oposto's model list flags this on the scene.
Two common-sense rules, whatever the model: only use a photo of someone who has given you permission, and use a sharp, front-facing, well-lit photo. Several photos of the same person help the model keep them consistent from scene to scene.
What lengths, formats and quality levels does it support?
| What ByteDance publishes | In Oposto | |
|---|---|---|
| Clip length | 4 to 30 s, can be extended twice | 4 to 30 s per storyboard scene, up to 15 s for a shot made in the studio |
| References | Up to 50 (images, video, audio) | The photos and videos you attach to the scene |
| Quality | 480p and 720p in the listed plans | 480p, 720p or 1080p, chosen per scene |
| Aspect ratios | Not stated on the pages reviewed | Vertical 9:16, square 1:1, landscape 16:9 |
| Sound | Generated with the picture | Sound created with the picture |
For a Reel, a TikTok or a Short, go vertical 9:16. Our image and video size cheat sheet lists the sizes platform by platform. As for quality, 720p is often enough for a video watched on a phone; save 1080p for the shots where detail really matters, because it costs a lot more.
How much does Seedance 2.5 cost?
ByteDance doesn't publish a per-second price on the pages reviewed. Its cloud service bills by tokens, and its plans are described in seconds of video: the smallest one advertises roughly 120 to 500 seconds of 480p video, depending on settings. The real price depends on quality, length and the references you use.
In Oposto you don't have to do that math. Each scene's panel shows the price of each quality level right next to the choice, and the storyboard gives the cost per scene before you start generating. Seedance 2.5 is still the most expensive model on the list, so keep it for the scenes that carry the video.
Seedance 2.5, Gemini Omni 1.2 or Kling 4.0: which one should you use?
All three make realistic videos with sound. They are not built for the same job.
| Seedance 2.5 | Gemini Omni 1.2 | Kling 4.0 | |
|---|---|---|---|
| Best at | The best-looking result, faithful to references | A short, simple scene | A long scene on a budget |
| Scene length in Oposto | 4 to 30 s | 3 to 15 s | 3 to 30 s |
| References | Photos, video and audio | Up to 6 images, short videos | Photos and video |
| Real person from a photo | Accepted in Oposto | Accepted | Accepted |
| Price | The most expensive | The cheapest of the three | Low |
| Oposto plan | Pro and up | From the free plan | Pro and up |
For Gemini Omni 1.2, Google describes clips of 3 to 15 seconds (10 seconds at most with a start and an end image), up to 6 reference images, and the ability to edit or extend a clip it created, up to 30 seconds in total. It doesn't take audio references. It is the model Oposto suggests by default for a simple scene, and the only one of the three on the free plan. Our Gemini Omni 1.2 article covers this version in detail.
Seedance 2.5 or Gemini Omni?
Seedance 2.5
- One continuous shot up to 30 s
- Several photos, videos and sounds to follow
- A real person to reproduce
- The opening scene, the product shot
Gemini Omni
- A short, simple scene
- A cutaway, a mood shot
- A tight budget
- The free plan
Kling 4.0, from Kuaishou, sits in between: in Oposto it makes scenes of 3 to 30 seconds in 720p or 1080p at a much lower price per second than Seedance 2.5. It is the right call for a long mood shot or a 20-second scene where the most polished look isn't essential.
Nothing forces you to use a single model in one video. A good split: Seedance 2.5 for the hook and the product shot, Gemini Omni for the cutaways. The first three seconds are the ones that deserve the best render.
How do you use Seedance 2.5 in Oposto?
You write what you want in the chat or in the video studio, and Oposto prepares the storyboard: the list of scenes, with what you see and hear in each one. That is where you pick the model.
Seedance 2.5 in an Oposto storyboard
- Describe the video Oposto splits it into scenes, with what you see and hear.
- Attach your references A photo of the person, the product, a video to follow.
- Pick Seedance 2.5 The list shows what the scene needs and which models fit.
- Pick the quality 480p, 720p or 1080p, with the price of each option.
- Create the scenes, then edit The edit puts the scenes together into one video.
Each scene shows its model and length at the top. When you open the model list, Oposto spells out what the scene needs (an attached photo, a reference video, a length), then states under each model whether it fits and why. Models that fit move to the top, so you don't have to remember which one accepts what.
Around the model, Oposto keeps its own rules: when someone speaks, the speech sets the length and the voice is never cut to fit a planned duration, and no app interface is drawn by the video model (screens are shown at an angle, blurred, or as a real screenshot). To write what gets said in the scene, a 30-second video script template helps you hold the pace. The full workflow is on the video studio page.
How do you get a good scene out of Seedance 2.5?
- One scene, one action. Describe what happens, in order, in plain sentences: who, where, what gesture, what framing.
- Sharp references. A blurry or dark photo gives a rough result. For a product, several angles beat a single shot.
- The right model in the right place. Keep Seedance 2.5 for the scenes that matter; cutaways work fine on a cheaper model.
- Review before you post. Check hands, visible text and branded objects. AI video on social media gives itself away mostly in those details.
Frequently asked questions
What is Seedance 2.5?
Seedance 2.5 is a ByteDance model that creates a video and its sound from text and references (images, video, audio). ByteDance announces up to 30 seconds in one generation and up to 50 references.
How long can a Seedance 2.5 video be?
ByteDance announces 30 seconds in a single generation, with two possible extensions. In Oposto, a Seedance 2.5 storyboard scene runs 4 to 30 seconds; a shot created or edited in the video studio is capped at 15 seconds.
Does Seedance 2.5 accept a photo of a real person?
In Oposto, yes: a Seedance 2.5 scene accepts a photo of a real person as a reference. Only use a photo of someone who has given you permission.
How much does Seedance 2.5 cost?
ByteDance doesn't publish a per-second price: its cloud service bills by tokens. In Oposto, the price of each quality level (480p, 720p, 1080p) is shown on the scene before you generate it.
Seedance 2.5 or Gemini Omni 1.2: which should I choose?
Seedance 2.5 for a scene that has to look great, run long or follow several references; Gemini Omni 1.2 for a simple 3 to 15 second scene, cheaper and available on Oposto's free plan.
Can I use Seedance 2.5 for free in Oposto?
No, Seedance 2.5 is available on the Pro plan and up. On the free plan, scenes are made with Gemini Omni.
Make your videos with Seedance 2.5 in Oposto
Oposto builds your video's storyboard from one sentence, lets you choose the model and quality of each scene, then edits it all together with voice, music and captions. Seedance 2.5 and Kling 4.0 are available on the Pro plan and up; on the free plan, scenes are made with Gemini Omni. You can start without paying.