Magnific adds 3D camera controls for Seedance 2.5 videos

• Camera direction becomes a reference input: magnific's bowling demo blocks a track, hold, and whip-pan before Seedance 2.5 generates the scene.
• The same system can capture a move in AR on a phone, according to magnific's walkthrough, and apply it across multiple shots in one take.
• Prompts still determine the people, place, and action. magnific's six-person demo uses the 3D setup for camera movement while the scene remains prompt-defined.
• Slow push-ins and creeping moves need prompt language as reinforcement, magnific's note says.

A bowling camera that follows the ball, stays through the strike, then snaps back to the player is the opening test in magnific's video. A second magnific demo sends the camera through six people around a table without losing a face.

Magnific's official Seedance 2.5 announcement frames 3D Motion Control as a way to set the camera path in space before generation.

The blocking workflow in magnific's demo has three moves:

1. Pin the camera at its starting position.
2. Move it through the desired path.
3. Pin it again, then hand that path to Seedance 2.5 as a motion reference.

The alternative is to record the move in AR with a phone. Magnific describes the resulting reference as usable for a multi-shot take, where the creator keeps the camera move and the model supplies the world; magnific's launch card says the control is available now.

A motion reference controls movement rather than content, according to magnific's explanation. The prompt still needs to specify the scene, otherwise the model fills those details itself.

That division is visible in the table example: Magnific Agent sets a six-person scene in 3D without conventional modeling, then the generation follows the camera through it. Magnific's Spaces documentation similarly describes its wider Video Generator as combining text prompts with optional visual references.

The 3D track does not replace every camera instruction. For slow pushes and creeps, magnific explicitly asks creators to repeat the move in the prompt in its note on the table shot.

CharaspowerAI takes the opposite starting point, preferring prompts that specify when the camera accelerates, slows, rotates, or changes scale. Higgsfield offers another route, saying in higgsfield_ai's Blender plug-in demo that locking motion in Blender is useful because camera movement remains one of AI video's hard problems.

Magnific's official announcement lists up to 50 references, 30-second generations, editing, extension, and 4K upscaling, with access through Magnific, Magnific MCP, Spaces, and the API.

The adjacent Magnific Desktop discussion adds four client-side details:

• Projects sync to a folder on the machine, according to techhalla.
• Tabs remain available the next day, techhalla wrote.
• The app notifies the user when a job completes, techhalla added.
• On Mac, it can be launched from Shortcuts, Spotlight, or Siri in techhalla's thread.