Modly sits at #9 on GitHub Trending today with +579 stars in a single day, taking it to 5,966 stars and 604 forks. It is a desktop application for Windows, Linux, and Apple Silicon macOS that turns a photo or a text prompt into a 3D mesh using open-source models that run entirely on your own GPU — no cloud round-trip, no per-generation billing. The repository opened on 17 March 2026 under the MIT license, built by a solo developer publishing as Lightning Pixel.

The part that separates Modly from a single-model wrapper is that it ships no model of its own. Generation capability arrives through an extension system, where each extension is a GitHub repository containing a manifest.json plus the runtime entry files its type requires — installed by pasting an HTTPS repo URL into the Models page. Five official extensions are published today, and they read as a tour of the open image-to-3D field: Hunyuan3D 2 Mini, Hunyuan3D 2 Mini Turbo, Hunyuan3D 2 Mini Fast, TripoSG, and Trellis2 GGUF. Swapping the generator is an install, not a fork, and third-party extensions install by the same path.

Generation is expressed as a node graph rather than a single button. The documented starting point is Image → Generate Mesh → Add to Scene, wired on the Workflows tab and then run from the Generate tab. Wiring is validated before a run, and this is a genuinely thoughtful detail: an invalid graph stays in place and surfaces inline and toast warnings instead of dropping the mesh you are already looking at. The v0.4.1 release on 16 July 2026 pushed the graph model further with While Loop nodes, For Each iterator nodes for image, text, and mesh, per-slot text inputs on extensions, and audio process outputs. Imported meshes can be smoothed and decimated in-app, with optimized results written back into the workspace.

There is also a stdlib-only CLI at tools/modly-cli/agent.py so agents and scripts can drive a running Modly app without the UI, emitting machine-readable JSON on stdout. generate --image ./input.png --output ./export.glb starts a workflow run, polls it, and exports the mesh, returning recovery metadata such as workflow-run status and workflow-run cancel in the response. The canonical commands are health, model, workflow-run, capability, and process-run; the docs are explicit that legacy, dev serve-api, and the experimental ComfyUI helpers are compatibility surfaces, not the agent contract, and that dev ensure-server starts only the FastAPI backend without proving the Electron bridge is ready.

The limits are real. macOS support is Apple Silicon only, there is no hosted fallback if your GPU cannot hold the model, and the app is still labelled Beta at v0.4.1 with 57 open issues. The most recent commit to main dates to 25 July 2026, so day-to-day development has been quieter than the star chart suggests. The MIT license also carries an unusual author-added clause: forks that ship as their own app must keep visible credit to Modly and Lightning Pixel in the UI or documentation, which is why GitHub's license detector reports the repository as "Other" rather than plain MIT.