Vision Prep

Preprocess large images and PDF pages before sending them to a vision model. Use when the task mentions reading a large image, a screenshot with small text, a PDF page, or "大图看不清", "PDF 里的图片", "识别 PDF", "图片文字太小". DeepSeek vision downscales each image to ~800x800 and rejects PDF input, so large images must be tiled and PDFs rasterized first.

znlgis Updated 79 repo stars

File contents

znlgis/my-opencode-deepseek-config commit 8b2844867a

Frequently asked questions

npx get-skill add znlgis/vision-prep-2