DreamCube

From the project README

Independent Modly integration for DreamCube, maintained by DrHepa. It accepts a front RGB image, generates or uses a front depth map, estimates sparse cubemap depth, runs DreamCube when requested, and returns images or a GLB mesh through Modly jobs; scene manifests are retained as auxiliary sidecars.

This repository is the Modly extension, not the official DreamCube project. DreamCube source code and model weights remain separate third-party assets. The pinned upstream source does not declare a source-code license; the model repositories declare their own weight licenses.

Features Internal auto-depth by default using Depth Anything V2 Small. Optional manual depth input for users who already have a front-view depth map. Setup status and logs written under .modly/setup/ for installation diagnostics.

Requirements Modly with local model-extension support. Python 3 and git available to the setup environment. A CUDA/PyTorch-compatible NVIDIA GPU for practical DreamCube generation. Disk space for the extension-local virtual environment, the DreamCube upstream checkout, model weights, and runtime caches.

The currently validated development lane is Linux ARM64 with Python 3.12 and NVIDIA CUDA. Other operating-system, architecture, Python, or GPU lanes should be treated as unverified until they are tested; setup reports an actionable error instead of silently pretending that an unsupported dependency lane is ready.