Magnific adds Seedance 2.5 with 3D motion control and a 4K pipeline
By Steven Van ·
The video model now runs on Magnific with 50 references, 30-second clips, and in-app editing, extending, and upscaling.
ByteDance's Seedance 2.5 just arrived on Magnific, the all-in-one creative platform formerly known as Freepik, and it did not show up alone. Magnific paired the model with a set of platform-exclusive additions: a brand new 3D motion control tool built in-house, support for up to 50 references per generation, native 30-second clips, and a single workflow that lets you edit, extend, and upscale results to 4K without leaving the app. For creators who have been juggling separate tools for generation, editing, and upscaling, this is the pitch: one model, one pipeline, start to finish.
What Seedance 2.5 actually changes
Seedance 2.5 is ByteDance's new flagship video model, unveiled at the Volcano Engine FORCE conference and rolled out at the end of July before landing on partner platforms this week. The headline capability is native 30-second generation in a single pass, roughly triple what most competing models produce in one shot, and the official Seedance page notes clips can be extended twice for longer narratives. The model is also noticeably better at reading reference videos: instead of just copying motion, it picks up intention, framing, and cinematic language from the material you feed it.
Reference capacity is the other big number. Seedance 2.5 accepts up to 50 multimodal inputs in one generation: images, video clips, and audio tracks together. That is enough to lock a character's face, a product, a location, a style, and a soundtrack all at once, which is exactly the kind of control commercial work demands and casual prompting cannot deliver.
The Magnific-only layer: 3D motion control
The piece Magnific is most proud of is not the model itself but what the team built around it. The new 3D motion control tool lets you define camera movement spatially instead of describing it in a prompt. You set the path, and the scene you configure becomes the reference the model follows. Magnific frames it as "the model, plus everything it was missing," and camera control has genuinely been the weakest part of prompt-driven video: the same words produce different moves on every run. Setting the move in 3D makes it repeatable.
Edit, extend, and upscale to 4K in one workflow
Generation is only half the job on most projects, and this is where Magnific's platform play shows. After Seedance 2.5 produces a clip, you can edit it, extend the shot, and push the result to 4K with Magnific's own upscaling technology, all inside the same workflow. Seedance 2.5 supports regional editing, meaning you can change part of a frame without regenerating the whole clip, so revision passes stop being full re-rolls. For client work, that turns a slot-machine process into something closer to an actual edit session.
Available everywhere Magnific runs, including MCP and API
Availability is broad from day one: Seedance 2.5 is live on the Magnific platform itself, plus MCP, Spaces, Plugins, and the API. The MCP support is worth flagging for anyone building agentic workflows, since it means AI assistants can call Seedance 2.5 generations directly. Teams that want the model inside existing pipelines can go through the API instead of the web app.
Magnific has been aggressive about same-week model integrations since the Freepik rebrand, and this release continues the pattern: take a strong base model, then differentiate on tooling, workflow, and distribution. With Seedance 2.5 also rolling out to competitors this week, the 3D motion control tool and the 4K pipeline are the reasons to run it here rather than anywhere else.
Sources: Magnific announcement on X, Magnific: Seedance 2.5, TechNode: ByteDance launches Seedance 2.5