WobblePic v0.9.7 is now available! This is a feature-packed release that introduces Pin Deformation, Segment Move, smoother mask boundaries, and broader GPU support. Here’s everything that’s new since v0.9.4.
New Features
Pin Deformation
The biggest addition in this release. During a wobble drag, press Right-click or P to pin the deformation in place. The image stays stretched even after you release the mouse, letting you build up multiple deformations across the image.
- Pin up to 4 points simultaneously
- Each pin is marked with a red dot — click it to release (with bounce animation)
- Click outside the mask to release all pins
- ESC releases all pins and segmentation instantly
- Pinned areas resist new wobble forces with gradient damping — the closer to a pin center, the more it holds its shape
Segment Move
Use Ctrl + Drag to move a segmented object around while the background stays fixed. The movement is handled through shader texture coordinate offsets, so it’s smooth and GPU-accelerated.
Box Prompt for Segmentation
In addition to clicking on an object, you can now drag a rectangle around it for more precise segmentation. During the drag, a semi-transparent overlay shows the selection area. Box prompts give SAM2 more context, resulting in better masks for complex scenes.
Improvements
Bilinear Mask Boundary Upscaling
SAM2’s decoder outputs a 256×256 logits map that needs to be upscaled to the full image resolution. Previously, nearest-neighbor upscaling produced visible staircase artifacts along mask edges. WobblePic now uses bilinear interpolation on the raw logits before thresholding, producing smooth, pixel-accurate boundaries. Read our deep dive on this technique for the full details.
ORT Model Support for Faster Startup
WobblePic now supports ONNX Runtime optimized (.ort) models in addition to standard .onnx files. ORT models have pre-applied graph optimizations that significantly reduce session loading time:
- Encoder loading: 67% faster (2,855ms → 933ms on GPU)
- Decoder loading: 30% faster (268ms → 187ms on GPU)
- Inference speed remains the same — the improvement is purely in startup time
If ORT files are present in the models/ directory, they are automatically preferred over ONNX files.
GPU Multi-Vendor Support (DirectML)
SAM2 inference has been migrated from CUDA (NVIDIA-only) to ONNX Runtime + DirectML, enabling GPU-accelerated segmentation on:
- AMD GPUs
- NVIDIA GPUs
- Intel GPUs
This means more users can enjoy fast AI segmentation without needing specific GPU hardware.
Code Refactoring
The codebase has been reorganized for better maintainability:
- Segmentation code moved to a dedicated
seg/directory - Utility functions extracted into separate modules
- Mouse event handling refactored for readability
Bug Fixes
- Fixed P key pin not working during drag due to imgui keyboard capture
- Fixed context menu appearing when left mouse button is held
- Fixed overlay ad showing on first image before SAM2 loading starts
- Fixed uninitialized timing variables in SegManager
- Removed dead code and unused pin methods
- Fixed max pin overflow — now releases all pins with bounce animation
System Requirements
Updated from previous versions:
- OS: Windows 10 or later (64-bit)
- GPU: OpenGL 3.3+ (AMD / NVIDIA / Intel with DirectML for SAM2)
- RAM: 4 GB minimum (8 GB recommended)
- Disk: ~600 MB (SAM2 model included)
Download
Download WobblePic v0.9.7 from GitHub Releases
If you’re upgrading from a previous version, simply run the new installer — your settings will be preserved.
What’s Next
We’re working on expanding the Gallery with more categories, improving the website with additional content, and exploring new interaction modes. Stay tuned!
Thank you for using WobblePic — now go pin and wobble something!