Higgsfield demos GPT-6 Astra building interactive 3D tools

• Interactive 3D is the point of Higgsfield’s Astra demos: the wardrobe demo swaps a look from a virtual rack, while the room-planner demo turns layouts into editable spaces.
• A Tokyo rail visualizer combines simulation with navigation, letting people follow trains, search routes or stations, and move from a single carriage to the whole city, per the rail-tracker demo.
• The strongest creative interaction comes from a wall-projected story that responds to hands and closes its virtual curtains on a clap, according to the theatre demo.
• Spline’s robot adds a more technical wrinkle: the walking demo says its solver detects collisions and corrects the robot’s pose as it walks.

Higgsfield’s Blender plugin guide describes an in-viewport set of image, video, 3D, animation and scene-building tools. Its MCP explainer says agent-generated assets consume credits, and a recent independent Astra roundup collected early reports of image-led Blender reconstructions becoming editable geometry.

The wardrobe demo makes each garment on a digital rack an input that becomes part of the next outfit, with character and clothing models made in Blender, as the wardrobe demo describes. The useful shift is from generating a single fashion render to retaining an editable wardrobe state between choices.

The room planner follows the same pattern. Higgsfield’s room-planner post describes layouts and lighting becoming editable 3D spaces; the clip moves from a marked-up room image to an editor where furniture can be repositioned.

Higgsfield’s Tokyo demo frames 3D as an information interface rather than a hero animation. Its stated controls are:

• Follow a train between stations.
• Search by train or station.
• Zoom from a carriage-level view to the citywide network.

The project also simulates train journeys, the tracker post says.

A projected bedtime-story theatre connects a scene to physical gestures. Higgsfield’s theatre post lists the stack as Astra, Higgsfield, Blender and TouchDesigner.

Its two controls are equally direct:

• Hand movements control the story.
• A clap closes the curtains.

Spline’s robot goes beyond a static model: the Spline demo says its walking solver detects collisions, then self-corrects the pose while crossing obstacles.

The toolmaker’s follow-up puts the process in plainer terms. Spline’s reply says an agent or MCP workflow still needs guidance about the intended form and behavior; in its spider example, that included mechanically feasible joint shapes. It also says the result did not come from one prompt.