Magnific releases depth-map workflow linking Claude Opus 5.5 to Seedance 2.5

• The workflow turns one reference clip into two inputs: Magnific's first step creates a depth map with Claude Opus 5.5 and the Magnific MCP, then Magnific's skier prompt feeds the source video and map to Seedance 2.5.
• Motion, framing, pacing, and blocking are treated as protected properties, while prompts replace characters, wardrobe, sets, and skies, as Magnific's jungle-film-crew prompt makes explicit.
• A second reference pattern separates jobs by asset: venturetwins' performer-swap recipe uses stills for identity and wardrobe, and the source video for performance, camera, lighting, and background.
• The MCP can also coordinate a larger chain of image, depth, animation, music, and delivery tasks, according to koldo2k's animation demo.

Magnific's official platform page says its MCP connects AI agents to its models and workflows. The sharpest sample turns a skier into an astronaut on a purple alien planet while holding the original camera movement, framing, pacing, and rider motion in Magnific's full edit prompt; another crosses four worlds from a fixed popcorn-box POV in Magnific's continuous-transition prompt.

Magnific's recipe starts with uploaded footage, then asks Claude Opus 5.5 through the MCP for a depth map. Magnific's first step supplies that map alongside the original clip to the next stage, and Magnific's edit prompt calls Seedance 2.5 with both video inputs.

That is a useful division of labor. The source supplies the moving shot, while the depth map gives the edit a spatial reading of it. Magnific directs people to its Motion Transfer post and Video Generator to try the workflow. Separately, Runway's Seedance 2.5 guide describes the model as reference-to-video, editing, and extension tooling that accepts text, images, video, and audio.

The prompt is structured less like a general style request than a shot brief:

1. Name the source asset: modify <<<video_1_1>>>.
2. Declare the invariant: keep everything else unchanged.
3. Specify every intended replacement, from the astronaut's helmet and board to dust, sky, trees, and planet surface.
4. Reassert the locks: preserve camera movement, framing, pacing, and the rider's motion exactly.

Magnific uses the same structure for a fashion edit. In Magnific's fashion prompt, the walking subject changes from a pink dress to denim, and a modern apartment becomes a Japanese temple, while both women's faces and movement, the photographer's sweater, camera movement, garden, and daylight stay fixed.

The jungle example changes a derelict brick interior into an overgrown stone temple and re-dresses the crew, yet keeps the same people, action, blocking, camera movement, framing, film camera, boom microphone, and clapperboard. The prompt assigns protection down to individual props rather than asking for an overall location makeover.

Creator venturetwins describes the same asset-role split for a performer replacement: still images establish identity and wardrobe, while the video provides motion, expressions, gestures, timing, interactions, cuts, framing, lighting, and background. He reports handling two replacement characters first, then running a second pass for the third.

Motion transfer is one small piece of the MCP pitch. koldo2k's detailed animation prompt asks for Seedream 5 Pro landscapes, GPT 2.5 transparent cutouts, depth maps, Kling 2.5 animation, and Lyria 3 music, with a list of generations and credit cost before expensive steps begin.

Its production constraints are unusually specific:

• Split every landscape into three depth layers and fill hidden areas.
• Use portal objects whose glass reveals the next world.
• Set a 20-second, 1920×1080, 30 fps loop, with music at 96 BPM for exactly eight bars.
• Deliver an interactive viewer with an AE-style timeline, plus an MP4 under 30 MB.

The requested artifact can expose variants, render, and download, koldo2k's follow-up says. koldo2k put the all-in-Magnific generation cost at 1,500 credits in a reply.

Magnific's separate Draft mode runs Seedance 2.5 previews at 480p before a selected take is regenerated at 1080p. It gives the motion-transfer setup a low-resolution ideation stage, distinct from the final render.