improved_vision_for_deepseek
Run the following command in DeepSeek Harness:
dsh plugin install zyh20041227/improved_vision_for_deepseek
Paste the following prompt into your AI chat to install this plugin:
To install this plugin in DeepSeek Harness, run the command: dsh plugin install zyh20041227/improved_vision_for_deepseek, or visit the full source at https://github.com/zyh20041227/improved_vision_for_deepseek.
About this plugin
DeepSeek's vision models often struggle with high-resolution images due to the 384-token limit, leading to missed details in small text on receipts, tables, or charts. The improved_vision_for_deepseek plugin tackles this by intelligently tiling images into a global overview, overlapping coverage tiles, and optional dense-region crops, ensuring each local area gets sufficient token budget to enhance recognition accuracy.
Its core capabilities include 100% geometric coverage, guaranteeing every source pixel is covered; overlapping seams so text or shapes crossing tile edges remain visible; content-aware cuts that shift horizontal seams to low-ink areas in document mode; and optional dense-region review for extra detail crops. Each tile includes traceable metadata like coordinates, role, and coverage for auditable output.
This plugin is perfect for users handling images with dense text or complex diagrams, such as data analysts, document digitization specialists, or researchers. It boosts DeepSeek model performance while maintaining stability through bounded batches and conservative token caps in the DeepSeek Harness environment. If you encounter recognition bottlenecks with vision models, this plugin significantly improves efficiency and reliability.
Screenshots
Use Cases
- Processing tables and dense text in scanned documents
- Recognizing small text labels in high-resolution diagrams
- Analyzing details in long screenshots or receipts
Best For
- document digitization engineers
- data analysts
- AI model developers
Related Plugins
ModLens is a vision plugin for DeepSeek Harness that gives text-only models sight by reading images pasted directly into chat, with zero-config setup and multiple vision engines.
On-demand vision for text-only DeepSeek Harness agents: built-in free keyless vision chain and 14 vision tools, routing image turns as tool calls to vision models with pixel fidelity, no Python needed, one-command install.
Give text-only models in DeepSeek Harness eyes, enabling image Q&A, long-screenshot OCR, UI restoration, and GUI visual tasks.
An armor-breaking plugin for DeepSeek: rewrites instruction priority, expands refusal-blocked phrases, persists across sessions, and shows a green active indicator.