Seedance 2.5
Seedance 2.5 is ByteDance Seed's current audio-video joint generation model. The official page says it is built for 30-second storytelling with precise reference control and stronger editing. You can generate up to 30 seconds in one take and extend twice for a longer story. Seedance 2.0 remains the in-app teaser inside Doubao. 2.5 is the research-site lead: smoother motion, more realistic visuals, and tools aimed at professional production.
Key Features
- 30 seconds, extend twice: One generation up to 30s, then two extensions for a longer narrative.
- Joint audio-video: Continues the Seedance line of audio-visual generation rather than silent clips only.
- Smarter references: Official copy says it reads intention, framing, and cinematic language from reference video, not only motion transfer.
- Pro editing: White-model control, green-screen editing, camera movement, and performance blocking are listed for complex production.
- Try / API / compare: The page offers Try Now, Compare Now, and Get API (on the 2.0 sibling page; confirm the live 2.5 CTA).
Use Cases
- Short-form story shots that need sound and a second extension instead of a 4-15s clip.
- Reference-driven brand film when a previous take should teach framing, not only pixels.
- Green-screen and white-model control for production teams that already composite.
Pricing
The Seedance 2.5 page we fetched does not publish a public per-second table. Consumer Doubao currently advertises Seedance 2.0 as free after login. API access is a separate Volcengine / Seed path. Do not invent a USD rate. Check the live Try/API buttons and Doubao's in-app meter.
Getting Started
- Read seed.bytedance.com/en/seedance2_5.
- If you only need a consumer trial, log into Doubao and see whether the picker still says 2.0 or already 2.5.
- For API work, follow the page's Get API / Volcengine path and confirm the model ID.
- Start with a 30-second take before you spend both extensions.
Frequently Asked Questions
Is this Seedance 2.0?
No. 2.0 is the Doubao homepage promo. 2.5 is the Seed site's next-generation 30-second model. Nav also lists 2.5 as current.
Open weights?
The public 2.5 page does not promise downloadable weights. Do not assume a local checkpoint.
Versus Veo or H3?
Veo 3 is Google's cinematic family (now 3.1, 1080p/4K). MiniMax H3 publishes $0.08-$0.13/s API rates for 768P/2K up to 15s. Seedance 2.5's published hook is 30 seconds plus two extends.
Alternatives
- Doubao: Consumer app currently bundling Seedance 2.0.
- Veo 3: Google DeepMind cinematic video.
- MiniMax H3: Pay-as-you-go 2K stereo, shorter clips.
Tips
- Keep 2.0 vs 2.5 straight in internal docs. Doubao and Seed are not always on the same number.
- Use extensions only after the first 30s works. Official story is generate, then extend twice.
- Recheck the Seed nav. 2.5 can be replaced the same way 2.0 was.
Conclusion
Seedance 2.5 is the ByteDance Seed video model to evaluate in 2026 when you need 30-second audio-visual stories and production edits. Start on the official 2.5 page, then confirm whether Doubao's free teaser has caught up.
Comments
No comments yet. Be the first to comment!
Related Tools
FLUX 3
bfl.ai/models/flux-3
Black Forest Labs multimodal model for video, audio, upcoming images, and action-prediction, with up to 20-second clips, native audio, and draft mode.
MiniMax H3
www.minimax.io/blog/minimax-h3
MiniMax open multimodal video model (July 31, 2026): text/image/video/audio in, 768P or 2K out, 4-15s clips, native stereo, pay-as-you-go API.
Veo 3
deepmind.google/models/veo
Google DeepMind's Veo video model, now presented as Veo 3.1: cinematic video with native audio, plus 1080p and 4K output on the official model page.
Related Insights
What Locks Codex Is the Picker, Not the Models
You already pay for OpenCode Go, Grok, and Z.ai, but the Codex picker still shows mostly GPT. The community project codex-router does not teach another install ritual. It puts subscriptions you already bought back into the selector, keeps keys on the machine, and leaves native GPT alone.
Seven AI Coding CLIs, Six Months: No Matter How Strong the Model, Work Needs Supervision
Claude Code, Codex, opencode, pi, omp and DeepSeek Harness all have personalities. After six months of deep use I run a division of labor: pi for the fastest cheapest reviews, omp for complex PRs, DeepSeek Harness on V4 Flash for high-frequency low-cost review, and Claude Code, Qoder and Cursor for writing. No matter how strong the model, work needs supervision — ideally from an independent third party.
After I Connected Obsidian to OpenClaw, It Started Helping Me Make Decisions
Once Obsidian stopped being just a place to store notes and started working with OpenClaw, it began helping me organize context, connect information, and improve real decisions.