Seedance 2.5

• 🎬 ByteDance's flagship image-to-video model, animating a still into cinematic motion
• 📏 Up to 30 seconds in a single generation, extendable twice
• 🆕 Joint audio-video generation rather than video first, sound later
• 🎯 Precise reference control plus consistency-preserving editing, per ByteDance
• ⚡ Smoother, more consistent motion and more realistic visuals than 2.0
• 🔧 Sibling endpoints cover text-to-video and multi-reference-to-video
• 🏢 Released in 2026 by ByteDance's Seed research team
• 👁️ Input image sets the opening frame; prompt directs motion and camera

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 image-to-video endpoint of ByteDance's newest Seedance generation, developed by the company's Seed research group. You supply a still image as the opening frame plus a text prompt describing action, camera work, and atmosphere; the model extends that frame into a continuous shot. ByteDance describes Seedance 2.5 as a next-generation audio-video joint generation model built for 30-second storytelling with precise reference control and editing capabilities.

The clearest step up from Seedance 2.0 is duration and continuity: ByteDance says clips can be created up to 30 seconds in a single generation, with the option to extend twice for longer, more complete narratives, and states that motion is smoother and more consistent while visuals are more realistic. Because audio and video are generated jointly rather than in sequence, synchronization is handled inside one pass rather than bolted on afterwards. Compared with earlier tiers such as Seedance 1.5 Pro and the lighter Seedance 2.0 Mini, the 2.5 flagship is aimed at longer, more directed shots rather than short clips.

Within the 2.5 family, this listing is the image-conditioned variant. Seedance 2.5 text-to-video starts from a prompt alone, while Seedance 2.5 R2V accepts multiple reference assets for tighter control over character, style, and rhythm.

Typical uses include animating product photos, portraits, concept art, or frames produced by ByteDance's image models such as Seedream V5 Pro, where preserving the subject's appearance across the whole clip matters most. ByteDance has not published a technical report for the model, so architectural detail beyond its launch materials remains limited.