Short updates on the mesh generation path, editor workflow, deploy readiness, and what changed as the product moved to the new high-fidelity pipeline.
Product direction
Fidelity first, artist imitation later
ClearMesh is now shaped around high-fidelity mesh generation rather than artist-specific style transfer. The new generation stack gives the product a stronger geometry baseline, and feed-forward editing keeps the user in a chat-first loop instead of a traditional DCC workflow.
The GPU path produced a real GLB in the CUDA smoke
The successful H100 smoke generated a dense GLB, passed fast mesh QC, and produced a nonblank Blender render. That makes the GPU path a real product integration target, not only a planned capability in the UI.
The Studio side panel now tracks selected objects, local masks, transforms, snapping, axis constraints, material previews, shading mode, and recent editor operations. Those details are sent through the edit request so the backend can preserve the user’s exact edit intent.
The mesh library now reviews versions and artifacts
The Library page can filter by mode, project, artifact type, and latest versions. The detail panel shows a version timeline, stored artifacts, GLB downloads, and a Studio handoff for continuing an edit.
The production gate checks public pages, Studio assets, `/health`, authenticated inference config, generation bakeoff evidence, and both bridge and app smoke reports. A production pass should mean the web app, GPU bridge, storage, auth, billing, mesh quality gate, and artifacts were all exercised.