Latest Version
divshot v2.1.0.pre.2
Released: March 11, 2026
What's New:
- Improving gaussian splat training quality
- Support image file lists drag into sceneview to start training
- Support export gaussian splats data during training
- Improving dataset loader
- Support saving training checkpoint data so that training can be resumed from the last saved point next time
- Support auto-save (when auto-save is enabled in preferences)
System Requirements
Operating System
Windows 10 or later (64-bit)
GPU
NVIDIA GPU with Compute Capability 6.0+ (GTX 1060 or higher) or AMD GPU/Intel GPU with Vulkan support
VRAM
8GB or more recommended
RAM
16GB or more recommended
Previous Versions
Looking for an older version? Download previous releases below.
v2.1.0.pre.1
March 11, 2026
- fix 3dgs export
- update incremental mapper sparse reconstruction and add global mapper
- optimize large dataset loader
- fix sparse point cloud ply loader
- support camera transforms and point cloud export
v2.0.2
January 20, 2026
- fix lots of bugs
- vulkan backend support fisheye model training
- Supports dragging camera pose files and sparse point cloud files with arbitrary names directly into the training dialog
- Improved the training preview rendering method, greatly enhancing the training preview quality
- Optimized the CUDA backend ADC+ densification strategy
- CUDA backend is now disabled by default, Mip Filter is enabled by default
v1.3.1
December 15, 2025
- fix video frame extract
- support more video encode for video frame extract
- use pinhole camera on default settings
v1.3.0
November 28, 2025
- Added fisheye camera Gaussian training and distorted Pinhole camera training, currently supporting 4 camera models: SIMPLE_PINHOLE, PINHOLE, OPENCV_PINHOLE, OPENCV_FISHEYE
- Added Vulkan Mesh Shader rendering for 3DGS to achieve more accurate 3DGS projection calculations, Vulkan fisheye camera support, and depth of field rendering effects
- Further improved training quality of SplatADC+ strategy, added a new parameter GrowFraction in the UI to control the number of Gaussians grown during each training iteration
- Further improve gpu memory using for large number images using sparse grad
- Fixed KeyFrame Panel timeline occlusion issue in non-black mode
- Fixed font size being too small on 4K screen resolution
- Fixed pure point cloud PLY file import rendering issue
- Fixed some program startup crash issues
v1.2.3
November 10, 2025
- Improved gsplat Vulkan rendering quality and performance (improved GPU sort through indirect dispatch, more accurate splat quad range calculation, reducing flicker when camera moves)
- Increased training image cache size, now supports up to 24GB of in-memory image data for training
- Added sog format support for gsplat storage and loading, this format can reduce storage space by 20x
- Updated spz data format loading and storage, now supports loading and storing gsplat data in spz format
- Editor camera properties are now displayed and adjustable in the inspector panel
- Modified SFM sparse reconstruction parameter optimization, slightly improving reconstruction speed
v1.2.2
October 25, 2025
- Fix the issue of the global pause training status on the toolbar
- Fix the editing issue of the Gaussian model that does not enable anti-aliasing in the rings mode
By downloading, you agree to our Terms and Conditions and Privacy Policy