picturereader

picturereader addresses a simple but frustrating limitation: pure text models cannot directly see images or read visual documents. For DeepSeek and other text-only models, pasted images may trigger unsupported-content errors or become unusable in conversation. The plugin creates a visual twin adapter for selected models, enabling native thumbnails in DeepSeek Harness while intercepting image blocks and converting them into local file paths, pixel-level evidence, and text guidance that the model can understand.

Its core strength is a complete local pipeline for reading and processing images. It provides scanning, cropping, pixel sampling, palette extraction, comparison, batch validation, and OCR with multiple engines, and can turn PDF, Word, Excel, and PowerPoint files into page images for analysis. It also includes document-to-image conversion and a local image editing tool with CPU-only actions such as resizing, rotating, filters, watermarking, compositing, background removal, stitching, and perspective correction. Three routing modes—privacy, smart, and strict—control whether to keep everything local or optionally call an external vision API for deeper semantic understanding.

This plugin is a good fit for users who want text-only models to handle screenshots, scans, contracts, tables, charts, and image-heavy workflows without switching to a fully multimodal model. It is especially useful for privacy-conscious users who want local analysis by default, while still leaving room for optional external vision services when more complex semantic understanding is needed.