Seedance 2.5

• 🏢 ByteDance Seed's flagship text-to-video model, released August 2026.
• 🆕 Succeeds Seedance 2.0 at the top of the family.
• 💬 Joint audio-video generation: dialogue, ambience and score arrive together.
• 📏 Built for roughly 30-second storytelling in one generation.
• 🎯 Reference-driven control for consistent characters, props and style.
• 🔧 Editing and extension are part of the same model.
• 🌐 Sibling endpoints cover image-to-video and reference-to-video inputs.
• 👁️ Prompt in, cinematic motion and scene consistency out.

ByteDance

ByteDance is a Chinese internet technology company headquartered in Beijing's Haidian district, with an associated variable-interest entity incorporated in the Cayman Islands. Best known as a consumer internet operator at enormous scale, it has also become a…

Seedance 2.5 is the text-to-video entry point to ByteDance Seed's flagship video family. ByteDance describes it as a next-generation audio-video joint generation model aimed at roughly 30-second storytelling, with precise reference-based control and built-in editing capabilities. In this catalog it is dated August 2026 and takes over from Seedance 2.0 as the family's top tier.

The generational shift is mainly about length and control. Earlier Seedance releases were oriented toward short cinematic shots; 2.5 is designed around multi-shot narrative clips of about half a minute, with reference inputs used to hold characters, wardrobe and props steady across a take, and with editing and extension handled inside the same model rather than by separate tooling. Audio is produced jointly with the visuals rather than dubbed on afterwards, so speech, ambience and music line up with what is on screen.

Seedance 2.5 ships as a set of matched endpoints. This one takes text prompts, while Seedance 2.5 image-to-video animates a starting frame and Seedance 2.5 R2V conditions on reference material, all sharing the same underlying model.

For cheaper or faster iteration, the family also includes Seedance 2.0 Mini and Seedance 2.0 Fast, plus the earlier Seedance 1.5 Pro. For still images feeding image-conditioned pipelines, ByteDance's Seedream V5 Pro is the companion model.