How to Use Seedance: ByteDance's AI Video Model, Step by Step

To use Seedance, open a platform that runs it, such as Dreamina or CapCut outside China, choose Seedance as the video model, then describe a shot in text or upload an image and describe how it should move. Seedance 2.5, the current version, generates up to 30 seconds of video with sound in one pass. And no, it is not open source: ByteDance keeps the model proprietary and offers it only through its own apps, its API and licensed partners.

That’s the short answer. Below is how each route works, how to get cleaner results, and how Seedance fits a YouTube video that runs for ten minutes instead of ten seconds.

What is Seedance?

Seedance is ByteDance’s AI video generation model, developed by the company’s Seed research team. It generates video from text, images, video clips and audio references, and recent versions create the soundtrack in the same pass as the picture.

Three versions matter:

• Seedance 1.0 launched in June 2025 as ByteDance’s first Seedance release.
• Seedance 2.0 arrived on February 12, 2026. It produces multi-shot clips of up to 15 seconds with stereo audio, and accepts up to 9 images, 3 video clips and 3 audio clips as references alongside your written instructions.
• Seedance 2.5 launched on July 31, 2026. It generates up to 30 seconds per generation, takes up to 50 multimodal references at once (30 images, 10 videos and 10 audio clips), and adds timestamp-level editing, so you can change one part of a clip without regenerating the whole thing.

Dreamina markets Seedance 2.5 as producing video at up to 4K. In practice, the resolution you get depends on the platform and plan you use it through.

Is Seedance open source?

No. Seedance is a proprietary model. ByteDance has published no model weights, no training code and no open license for Seedance 1.0, 2.0 or 2.5, so you can’t download it or run it on your own GPU.

What confuses people is the ecosystem around it. There are official client SDKs for calling the API, community projects on GitHub, and plenty of websites with “Seedance” in the name. None of those are the model. If a site offers a “Seedance download,” treat it with suspicion.

If you specifically need an open-weight video model you can run locally, look at releases like Wan 2.2 or LTX instead. Our guide to AI video generation models compares the open and closed options side by side.

Where can you use Seedance?

You can use Seedance through ByteDance’s own apps, its developer API, or a third-party platform that licenses it.

Where	What to know
Dreamina	ByteDance’s AI creation platform, and the first place Seedance 2.5 launched globally
CapCut	Seedance runs in CapCut’s AI video tools, including Video Studio. Seedance 2.0 reached US CapCut users in April 2026
Jimeng AI and Doubao	ByteDance’s apps in China. Select Seedance 2.5 under Video Generation
BytePlus ModelArk	ByteDance’s API for developers and businesses
Third-party platforms	API hosts such as fal, and products like TubeGen, run Seedance inside their own tools

Availability varies by version and region. Seedance 2.5 on CapCut and Dreamina began rolling out to subscribers aged over 16 in Europe, Asia, the Middle East and South America, so US users may still see 2.0 in some places. Access through ByteDance’s apps generally needs a paid plan. If you’re working inside TubeGen, the Seedance model page covers exactly what it costs and does there.

How do you use Seedance for text-to-video?

Pick Seedance as your model, write a description of the shot, set the aspect ratio and length, and generate. The steps look similar on every platform:

1. Open the video generator in Dreamina, CapCut or your platform of choice, and select Seedance.
2. Set the shape first. 16:9 for a YouTube scene, 9:16 for Shorts. CapCut launched Seedance 2.0 with six aspect ratios to choose from.
3. Describe the scene, then the action, then the camera. Who or what is on screen, where, what happens, and how the camera behaves. A named camera move, such as a slow dolly-in or a static wide shot, keeps the clip from drifting.
4. Mention the sound if it matters. Seedance generates audio with the video, so rain, crowd noise or a specific atmosphere can be part of the request.
5. Generate and review. Watch with the audio on. If one element is wrong in a 2.5 clip, use the editing tools to change that moment rather than starting over.

Text-to-video is fastest for exploring ideas. It gives you less control over exactly how things look, which is why image-to-video is usually the better route for anything that needs to match.

How do you use Seedance for image-to-video?

Upload or generate a still image, select it as the starting frame, and describe only the motion. The image fixes the look, so your words should cover what moves and how.

1. Start with a strong still. Composition, lighting and style all carry through, so fix them in the image before animating.
2. Add the image as your first frame or reference. Seedance 2.0 and 2.5 also accept extra reference images for characters, objects or locations you want held steady.
3. Describe the motion plainly. “Snow falling, the lantern flickers, slow push-in toward the door” works. Re-describing what’s already in the image mostly adds noise.
4. Avoid photos of real people. CapCut restricted making videos from images or videos with real faces, and ByteDance requires identity verification or legal authorization for real human portraits used as references. Generated characters sidestep this entirely.
5. Generate, review, adjust. Keep the clip if the motion works, and refine specific moments with the editing tools on 2.5.

Clips made in CapCut carry invisible watermarks, and for Seedance 2.5 CapCut also adds C2PA Content Credentials and visible AI labels. That’s worth knowing if you plan to publish the clips.

Tips for better Seedance results

Most weak Seedance clips come from overloading a single generation. These habits fix most of it:

• One main action per clip. Build a sequence from several short, clear shots rather than one busy one.
• Lock the look with an image. Text-to-video reinterprets your subject every time. A starting frame doesn’t.
• Use references for continuity. A character reference keeps the same face across shots, which matters in any series.
• Name the camera. An unspecified camera tends to wander.
• Keep motion purposeful. A slow push-in for tension, drifting particles for atmosphere. Constant dramatic movement reads as noise.
• Plan the audio layer. Generated sound is great inside a clip, but narration and music have to run across many clips.

How do you use Seedance for YouTube videos?

Seedance gives you shots, and a YouTube video needs dozens of them timed to a script. Generating each clip in Dreamina, downloading it and dragging it onto a timeline works for a Short, but it gets slow fast for a ten-minute narrated video.

TubeGen runs Seedance as its default animation model, inside the pipeline that builds the rest of the video. The flow looks like this:

1. Write or paste the script, then generate the voiceover in any of 8 languages.
2. Generate the scene images. TubeGen’s image generator reads the voiceover, splits it into scenes and creates an image for each one in your chosen art style.
3. Pick the scenes to animate. Seedance turns each one into a motion clip at 720p or 1080p. It shows up as “Default — balanced” in the animation quality picker, and it’s the cheapest way to animate in TubeGen.
4. Adjust a clip if needed. Describe the change in your own words, such as making a scene daytime, and it’s applied to the existing clip with the rest of the motion left intact.
5. Finish in the editor, where every scene, animated or still, is already timed to the narration, with music scored underneath.

Seedance clips in TubeGen have a minimum of about 3 seconds and run up to 8, which suits the pace of narrated video. How much you can animate per video depends on your plan: 1 minute on Starter, 10 minutes on Pro and unlimited on Premium. Other tiers are there for scenes that need more, such as Pixverse on the High tier for stronger motion. The animation feature page covers how animated and still scenes mix.

For a broader walkthrough on animated videos, including where motion actually earns its place, see our guide on how to create animations for YouTube.

The short version

Seedance is ByteDance’s closed-source video model, and the current 2.5 version makes up to 30-second clips with sound from text, images and references. Use it through Dreamina, CapCut, ByteDance’s API or a platform that licenses it, lead with an image when the look matters, and keep each clip to one clear action. When the goal is a full YouTube video, run it inside a pipeline that handles the script, voice and edit around it. See TubeGen’s plans to use Seedance that way.