InVideo AI adds ByteDance's Seedance 2.5 video model
By Steven Van ·
The model generates native 30-second clips in one pass and accepts up to 50 reference inputs.
Seedance 2.5 is now live on InVideo AI. The announcement was a single line, but the model behind it is not: ByteDance's newest flagship generates native 30-second videos in one pass, accepts up to 50 reference inputs, and adds region-level editing. For invideo's audience of marketers, educators, and small businesses, it is the strongest generative video engine yet added to a platform built around turning plain-language prompts into finished videos.
A frontier model inside an agent-driven editor
invideo's pitch has always been different from raw generation playgrounds. You describe the video you want, and the platform's AI handles scripting, footage, music, transitions, and assembly, with edits made through follow-up text commands or a traditional timeline. Seedance 2.5 joins a roster of more than 200 image, video, audio, and music models on the platform, alongside Seedance 2.0, Veo 3.1, Kling 3.0, and ElevenLabs music. In that context, the new model is not a separate tool to learn. It becomes another engine the invideo agent can direct on your behalf when a scene calls for generated footage.
What Seedance 2.5 brings to the table
The upgrade over Seedance 2.0 is substantial. Clip length doubles from 15 to 30 seconds, rendered as one continuous pass that holds character identity, lighting, and camera movement stable from first frame to last. Reference capacity grows from 15 inputs to 50 across images, video clips, and audio, so recurring characters and products can stay consistent across generations. Prompt adherence improves by roughly 20 percent, and region editing lets you fix part of a frame without re-rendering the whole clip. invideo's model page also highlights 3D white-model input, a control method where you block out a shot's composition with untextured geometry before the model renders it.
Why this matters for prompt-to-video workflows
A 30-second single-pass generation maps almost exactly onto the length of a typical ad spot or social clip, which is invideo's home turf. Until now, generated segments inside longer edits had to be short and carefully stitched. With Seedance 2.5 in the stack, a single generated clip can carry an entire short-form deliverable, and the platform's text-command editing pairs naturally with the model's region-level fixes. If you already use invideo for prompt-driven production, the practical move is simple: ask for Seedance 2.5 in your next project brief and compare it against the models you have been defaulting to.
Sources: invideo announcement on X, invideo Seedance 2.5 model page, HowAIWorks: Seedance 2.5 launch