OOOSplat 0.5.0: Auto-Optimization, Bridge Frame Interpolation and Re-Shooting

OOOSplat 0.5.0 is out — a local tool for turning video and photos into 3D Gaussian Splatting. The release adds automatic parameter optimization, bridge frame interpolation when coverage is insufficient, VRAM-aware training, and incremental capture for already finished projects. Details are in the release description on GitHub.

What it means

In our reference, OOOSplat is open source under the Apache-2.0 license, and the developer is ooolab. As described by the author, the application runs on Apple Silicon and Windows, and the entire process — frame preparation, camera reconstruction, training and PLY export — runs locally on CPU and GPU, without uploading to the cloud. The React interface via Tauri calls the Rust backend directly, with no remote services. What's new in 0.5.0 concerns primarily quality and stability on weak hardware. Auto-optimization itself plans the COLMAP and Brush parameters for the hardware: fast takes 6 FPS, — 8, — 12. If the initial video reconstruction does not cover the scene, the tool adds bridge frames within the budget, and on failure safely rolls back to the original result. For photo sequences, bridge interpolation is not applied — all frames are preserved. Worth noting separately is VRAM-dependent Brush training: it selects resolution and density for the available VRAM, and when there is not enough, it lowers the configuration once and retries. This directly addresses a common problem of local 3D tools — crashing on GPUs with a small amount of memory. Stability of GPU selection on Windows with a single NVIDIA card has also been improved. Incremental capture lets you extend a finished project with new frames without a full recompute: the COLMAP database, shared cameras and the sparse model are reused, and new frames are merely registered additionally. Transparent MOV and PNG are supported with Alpha preserved and COLMAP Mask generation. Incremental capture creates a separate project and does not touch the original. The preview is now rendered via WebGPU with automatic fallback to WebGL2 if the device does not support WebGPU. This should reduce jitter when rotating the camera and improve stability during long viewing sessions. The release covers several practical scenarios: re-shooting a scene to add details, working on machines with limited VRAM, and smoother previews of large models. For those already using OOOSplat, the update is worth considering first of all for crash protection and incremental capture.