Seedance 2.5 R2V

• 🏢 ByteDance's flagship video family, reference-to-video ("R2V") route, released August 2026.
• 🆕 Joint audio-video generation aimed at 30-second storytelling in a single pass.
• 📏 Clips can be extended twice for longer, more complete narratives.
• 🔧 R2V drives camera movement, pacing, blocking and shot scale from reference clips.
• 👁️ Improved lighting control derives physically plausible light direction, temperature and shadows.
• 🎯 Supports local video editing alongside generation.
• 🌐 ByteDance frames it as moving from clip-level outputs to full creative workflows.

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 R2V is the reference-to-video endpoint of ByteDance's Seedance 2.5 generation, sitting alongside the text-to-video Seedance 2.5 and image-to-video Seedance 2.5 routes. ByteDance describes the model as a next-generation audio-video joint generation system built for 30-second storytelling with precise reference control and editing capabilities, where a single generation can be extended twice for longer sequences.

R2V is the route that exposes the most explicit reference-based control of the three. Rather than a single prompt or hero image, it takes labelled reference assets and lets the prompt assign roles to each: ByteDance's own examples reference camera movement, pacing, shot-size transitions, subject trajectory and blocking from one clip while pulling materials, lighting, colour and atmosphere from a separate image.

Compared with Seedance 2.0 R2V and the lighter Seedance 2.0 Mini R2V, the emphasis shifts from short multi-shot clips to longer native takes with fewer stitched segments. ByteDance highlights smoother, more consistent motion and more realistic visuals, plus a specific upgrade in lighting control: spatial information from a reference (for example a clay render) is used to generate light that follows physical laws in direction, colour temperature, intensity and shadow projection. ByteDance frames 2.5 as a step from clip-level outputs toward complete creative workflows.

Practically, that means locking a character, set and palette across a longer take, then making targeted local edits. For still frames, the Seedream V5 Pro image models cover the same family's image side.