I lay down. A 0.23B open model sorted my screenshots folder.

Description and use

Screenshot Sloth uses the Florence‑2 model to OCR screenshots and automatically rename, sort, and flag files locally, without uploading data.

Problem

Users have large folders of screenshots with generic names, making it hard to locate specific items such as receipts, OTPs, or tickets.

Outcome

On a 12‑image demo set, Screenshot Sloth processed all images in 1 min 21 s, while manual sorting would take ~2 min. The tool provides a safe, offline workflow with undo capability.

Workflow

• Select folder via File System Access API
• Model downloads (≈350 MB) and runs in WebGPU/WebWorker
• OCR each image and extract text with bounding boxes
• Apply rules to determine filename, folder, and flags
• Show proposed plan to user
• User reviews and edits plan
• Apply changes: rename, move, flag, duplicate handling
• Log all moves to .screenshot‑sloth‑undo.json
• Provide Undo option

Evidence

Screenshot Sloth reads every screenshot in a folder with an open‑source vision model, inside your browser tab, and proposes a plan:

Source

The first run downloads the model once (about 350 MB, then cached). After that the demo goes like this:

Source

Machine time: 1 min 21 s to read all 12 (after the one‑time model download)

Source