Originally created by: BilgeGates
Originally owned by: BilgeGates
What problem does this solve?
Getting a position into ChessVision requires manually typing or pasting a FEN string. Users who have a board photo, a screenshot from a book, or an image saved from another app must read the FEN themselves or use a separate tool — a significant friction point that makes the app inaccessible for the majority of casual users who do not know FEN notation at all.
Proposed solution
Add an Open Folder action that lets the user pick a local image file (JPEG, PNG, WebP) containing a chessboard. The image is passed to a system-level vision model which detects the board, identifies pieces on all 64 squares, and returns a FEN string. The detected FEN is then loaded directly onto the board editor — same as if the user had typed it.
Interaction flow:
- User clicks "Open Folder" (toolbar or home page action)
- System file picker opens — user selects an image
- A loading indicator appears while the vision model processes the image
- On success — FEN is loaded onto the board; user can review and edit before saving
- On failure — a clear error message explains what went wrong (no board detected, ambiguous position, low image quality) with a prompt to try again
Scope for v6.3.0:
- Supported input: single image file (JPEG, PNG, WebP)
- Detection confidence threshold: if confidence is below a safe threshold, show the best guess with a warning rather than silently loading a wrong position
- Board orientation detection: if the model can infer which side is at the bottom, set the flip flag accordingly; otherwise default to white at the bottom
- Privacy: the image is processed entirely client-side or via a user-scoped API call — no image is stored or logged
Out of scope for this release:
- PDF or multi-page input
- Live camera / webcam capture (separate future issue)
- PGN extraction from game scores in images
Before submitting
- [x] I searched existing requests and this is not a duplicate.
