Seedance is ByteDance's video model family. Seedance 2.0 makes multi-shot clips with native audio from text, image, video and audio references; Seedance 2.5 extends that to 30-second generations with more references and extension. For personas, design a reference still first, then reuse it as an image reference in every clip.
Seedance AI video is ByteDance Seed's family of video models that generate clips with native audio from text, images, video and audio references, which makes it well suited to keeping an AI persona consistent across shots. Seedance 2.0 produces multi-shot clips of up to 15 seconds, and Seedance 2.5 extends a single generation to 30 seconds with more reference inputs. For persona creators, the useful workflow is reference first: lock the character in a still, then reuse it in every video.
Checked October 2026 against ByteDance Seed's Seedance 2.0 launch and Seedance 2.5 announcement, Dreamina's Seedance 2.0 page, the BytePlus ModelArk tutorial and Runway's developer API pricing.
Seedance is one of several video models persona creators use. For the full landscape, see our AI influencer video guide.
Seedance AI video: versions and what they do
| Version | Output | References | Where |
|---|---|---|---|
| Seedance 2.0 | Up to 15s multi-shot audio-video | Up to 9 images, 3 videos, 3 audio clips | Dreamina, BytePlus ModelArk |
| Seedance 2.5 | Up to 30s per generation, extendable | Up to 30 images, 10 videos, 10 audio clips | Jimeng AI, Doubao Pro; ModelArk API announced |
ByteDance describes Seedance 2.0 as a unified multimodal audio-video model: it can reference composition, motion, camera movement, effects and audio from your inputs, and outputs dual-channel audio aligned with the visuals. It also notes honestly that the model still has issues such as occasional audio distortion.
Seedance 2.0 and 2.5 in practice
- Up to 15-second multi-shot clips
- Native dual-channel audio
- 9 images, 3 videos, 3 audio references
- Available on Dreamina and ModelArk
- Good for short persona clips
- Up to 30 seconds per generation
- Extend a clip for longer stories
- 30 images, 10 videos, 10 audio references
- Timestamp-level editing
- Rolling out on Jimeng and Doubao first
For Reels and TikTok, 15 seconds is already a full post for many formats, so 2.0 covers most short persona content. The 2.5 features matter more for longer storytelling, where ByteDance says extension keeps characters, environments and pacing consistent across several generations.
A persona workflow: still first, motion second
- 1Design the persona still
A clean image of the character, for example with Seedream in Dreamina.
- 2Write a fixed persona description
Reuse the exact wording in every prompt.
- 3Add references
The persona still, plus any location image, motion clip or voice sample.
- 4Prompt one scene at a time
Subject, action, environment, camera move, lighting.
- 5Review and extend
Check consistency; on 2.5, extend rather than regenerate.
Dreamina's own guide suggests the same pattern: describe subject, action, environment, camera movement, lighting and style, upload your references with the + button, choose Seedance 2.0, then set duration, aspect ratio and resolution. For more on keeping a character stable across tools, see our consistent character guide.
Choosing references for a persona
Seedance accepts several kinds of reference, and each does a different job. ByteDance says the model can reference composition, motion, camera movement, visual effects and audio from your inputs, so think of each reference as an instruction rather than decoration.
- Persona image: the clean still of your character. Use the same one every time; switching between several faces invites drift.
- Location image: a recurring place, such as the persona's apartment or studio, so scenes feel like the same world.
- Motion clip: a short video showing the movement or camera move you want, such as a slow walk toward the camera.
- Audio clip: a voice sample or music style. Only use voices you have rights to, such as your own or a designed voice.
More references are not automatically better. Start with the persona image alone, add one more reference at a time, and keep the combinations that work in a simple notes file with the prompt used.
Native audio for persona clips
Native audio is one of Seedance's headline features: ByteDance says 2.0 outputs dual-channel audio and can generate background music, ambient effects and character voices aligned with the visuals. For a persona channel, that can cover ambient sound and music in one pass. If your persona has a fixed brand voice, you may still prefer to add voiceover separately so it sounds identical in every post; our guide to AI voice for video covers that route.
Longer stories with extension
Seedance 2.5 adds multi-round extension: you can append new shots to an existing clip while keeping characters, environments and pacing consistent, according to ByteDance. ByteDance's own example prompt asks the model to continue from a referenced video and generate another 30 seconds with the same characters, scene and style. For persona creators, that suits mini-series content, where each post continues the last, more than single standalone clips.
- 01Clip 1
Set the scene with the persona reference
- 02Extend
Continue the action, same characters
- 03Extend again
Resolve the story
- 04Edit
Trim, caption, add brand voice
Is Seedance AI free, and what does it cost?
It depends on where you use it. Consumer access runs through ByteDance platforms such as Dreamina, which has its own plans and credits; check its current plan page for free allowances. Developers can call Seedance through BytePlus ModelArk or third-party APIs. Runway's developer pricing, for example, lists Seedance models with per-second rates, at $0.01 per Runway API credit:
| Model on Runway API | Rate |
|---|---|
| seedance2 (480p/720p) | 36 credits per second |
| seedance2 (1080p) | 40 credits per second |
| seedance2_fast (480p/720p) | 29 credits per second |
| seedance2_5 (720p) | 30 credits per output second, plus input video seconds |
Source: Runway developer pricing, checked October 2026
Rates differ between platforms, so compare the per-second cost where you plan to generate. Our Runway Gen 4 guide and Hailuo AI video guide cover two alternatives with their own pricing.
Seedance AI vs Veo 3
Both generate video with native audio, so the decision comes down to access, cost per second where you work, and how well each holds your character. We have not run a side-by-side benchmark, so treat this as a checklist rather than a verdict.
- Same persona reference image on both
- Same three prompts: talking, walking, close-up
- Same aspect ratio and similar duration
- Note credits or cost per finished second
- Check audio quality and lip movement
- Check character drift across three clips
For the Google and OpenAI side of the market, see our Runway, Sora and Veo guide.
Consent and labelling
Multimodal references make it easy to borrow a real person's face or voice, which is exactly what not to do without consent. Build fictional personas or use your own likeness, and label realistic AI video where platforms require it: TikTok requires labels on AI-generated content with realistic images, audio or video, and Meta requires its AI-disclosure tool for photorealistic video or realistic-sounding audio that was digitally created or altered. For the full persona system, our AI Influencers program covers each step.
Seedance AI video: FAQ
What is Seedance AI video?
Seedance is ByteDance Seed's family of video generation models. Seedance 2.0 generates video with audio from text, image, audio and video inputs, and ByteDance says it supports multi-shot clips of up to 15 seconds with dual-channel audio. Seedance 2.5 extends single generations to 30 seconds. Checked October 2026.
What is new in Seedance 2.5?
ByteDance says Seedance 2.5 generates up to 30-second audio-video clips in one pass, supports multiple rounds of extension, accepts up to 30 images, 10 video clips and 10 audio clips as references, and adds timestamp-level editing. It launched on Jimeng AI and Doubao Pro, with API access via BytePlus ModelArk announced as coming soon.
Where can I use Seedance 2.0?
Dreamina, CapCut's creative platform, describes itself as the official place to access Seedance 2.0 and lets you upload up to 12 references (9 images, 3 videos, 3 audio clips). Developers can use it through BytePlus ModelArk, and some third-party platforms, including Runway's developer API, list Seedance models too.
Is Seedance AI free?
Access and credits depend on the platform you use, and terms change often. Check Dreamina's current plans for free credits and paid tiers, and the per-second rates on any API you use. Plan on testing your persona references with whatever free allowance is offered before paying.
How does Seedance compare with Veo 3?
Both generate video with native audio. Seedance's published strengths are multimodal references and longer single generations in 2.5; Veo is Google's model with its own pricing and access through Google products. We have not benchmarked them side by side, so run the same persona reference and prompt on each before choosing.
How do I keep a persona consistent in Seedance?
Use references. Generate a clean still of your character, for example with ByteDance's Seedream image model in Dreamina, then upload it as an image reference for every video, with the same persona description in your prompt. Seedance 2.5's extension feature is designed to keep characters and scenes consistent across longer sequences.
Persona clips with sound, story and consistency.
The AI Influencers program, included in All Access, covers personas, image and video workflows, voice and monetization, with the other three programs, live coaching and the private community in one subscription.
Free AI creator updates
Join the free Telegram channel for video model and pricing changes as they happen.