Send the GitHub link to your agent
Copy the message below into a local agent such as Codex, Claude Code or Cursor. The agent needs to read the repository and install the whole folder containing SKILL.md so the starter and helper files stay together. If your agent cannot install it from chat, follow the manual steps in the GitHub README for your client, then ask it to use kivo-white-model-video.
Install the kivo-white-model-video skill from https://github.com/miraga-kivo/kivo-white-model-video-skill and tell me when it is ready to use.
Describe one shot and review it in the browser
A white model uses simple, untextured shapes to show where people and objects are and how they move. Give your agent a location, an action, a camera move and a duration. It breaks the shot into timed beats, builds the figures and set in Three.js, and opens a HyperFrames timeline you can play and scrub. The starter includes follow shots, orbits, push-ins, crane moves and basic character actions such as walking, turning or sitting.
Use the same shot files for revisions. Ask for a wider opening, a slower push-in or a longer hold after the turn, then check those beats again. Once the shot works, export a reference MP4. Local preview and export require Node.js 22+, Chrome or Chromium with WebGL, and a full FFmpeg/FFprobe installation on the machine running the shot.
Use kivo-white-model-video to make a five-second shot. A person walks across a room toward a closed door, stops and turns toward it. The camera slowly pushes in. Give me a local preview with the key timestamps.
Use the approved MP4 as a Seedance 2.5 reference
Connect Kivo Studio MCP when you are ready to upload. Tell the agent which project and revision to use. Once the video is ready in that project’s Asset Library, bind it to a saved video prompt. Tell Seedance to follow the clip for camera movement, framing, actor trajectory and timing. Describe the final wardrobe, lighting and materials in the prompt, or attach separate image references for those details.
Ask for the cost of a single Draft before submitting. Local preview and export can run without a Kivo account. Kivo upload needs project access, and paid generation needs the project payer’s approval.
What carried through in our five-second test
In the door shot, Seedance followed the slow push-in, the left-to-right walk and the final turn toward the door. The actor stopped closer to the door, and the lighting shifted in the generated Draft. Reviewing the opening, walk and final hold side by side made it easy to see what carried through.
A local blockout lets you settle framing and timing before spending credits. Keeping the scene files means you can move the camera or adjust a beat and export another reference from the same scene. Seedance still interprets the reference, so check the Draft before committing to a final.
Kivo Skill and the Higgsfield Blender plugin
Higgsfield’s Blender plugin works inside the Blender viewport. Its Scene Builder places editable geometry and lights into the open scene. It also offers rigged character animation and Seedance 2.5 generation from the viewport, which fits teams already modeling or animating in Blender.
The Kivo Skill lets an agent create an editable Three.js shot and a HyperFrames browser preview without a Blender installation. Its output is the shot source and a portable MP4 that you can review locally and hand off to Kivo. This fits a quick test of an entrance, a camera move or a timing change when the shot needs only simple geometry. For a complex rig or a set that already lives in Blender, the Blender viewport gives you more direct editing controls.