Inspiration
Preparing slides, reports, teaching materials, and design references often requires extracting many small visuals from larger screenshots. General-purpose image editors are powerful, but repeating open-crop-save for every asset is slow. FrameCut was created to make this specific workflow direct and dependable.
What it does
FrameCut turns screenshots, slides, posters, and long images into organized visual assets. It lets users:
- Queue multiple images for continuous processing.
- Automatically detect regions likely to contain icons, illustrations, arrows, or pictures.
- Create, move, resize, delete, and classify crop regions precisely on a canvas.
- Preserve each image's regions, zoom position, and undo/redo history while switching through the queue.
- Cycle through nested or overlapping crop regions by clicking the same point again.
- Export full images and icons into organized subfolders at source resolution.
- Work entirely offline without accounts, API keys, telemetry, or uploads.
How we built it
FrameCut is a Windows desktop application built with Python, PySide6, OpenCV, NumPy, Pillow, and PyInstaller. Codex powered by GPT-5.6 was used throughout OpenAI Build Week to inspect the existing interaction model, redesign the interface, implement state-based undo/redo, build the multi-image document queue, solve overlapping-region selection, and verify the packaged Windows application.
The Build Week version was developed on a dedicated branch with dated source changes and a reproducible Windows package, making the new work distinguishable from the pre-existing utility.
Challenges we ran into
The hardest interaction problem was selecting a small crop region nested inside a larger one. A conventional hit test repeatedly selected the same top region. FrameCut now records every region under the pointer and cycles through them when the user clicks the same point again. A hover hint makes this interaction discoverable.
The second challenge was preserving complete editing state across multiple images. Each queue item now owns its crop geometry, labels, selected layer, zoom and pan state, and undo/redo stacks, so switching images does not destroy work.
Accomplishments that we're proud of
- A coherent desktop workflow rather than a one-off cropping script.
- Fast local startup with a reproducible Windows package.
- Per-image undo/redo and queue state with no cloud dependency.
- A practical solution for overlapping selection layers.
- Automated and packaged-app regression verification on Windows.
What we learned
Small workflow tools benefit more from predictable state and discoverable interactions than from adding many unrelated features. The Build Week work focused on product completeness: error recovery, multi-document continuity, clear visual hierarchy, and testability.
What's next for FrameCut
- Add an optional English interface.
- Run detection in the background with progress and cancellation for very large images.
- Save and resume complete queue projects.
- Add export presets for common slide, social, and documentation workflows.
Log in or sign up for Devpost to join the conversation.