Give text-only DeepSeek models on-demand vision
Leeminjing/dsh-eyes · model-adapters
To install "ready-available eyes" on a pure text large model (e.g. DeepSeek) in DeepSeek: the image pastes/annexes and stays backstage, with the model itself calling for a videomage tool to look at the map if needed (the bottom is any OpenAI compatible visual interface, default Qwen), just like the original model has multiple models.
★ 2GitHub stars
0Forks
2026-08-25Last updated
JavaScriptLanguage
MITLicense
Key features
- Lets map messages be stored by sending access and permanently
- Scratch the picture into a quote before sending it to the main model.
- Registers the view image tool, which is called by the main model at any time. The visual model extracts the description/OCR text and gives it to the main model to continue with its answers.
Requirements
- Two internal packaging pointsBecause Harness has no public extension point for "photocompetence judgement to send access" and "pre-dispatch photos", this plugin directly packages llm.resolveModelInfo with llm.streamWithRegistration.
- The side effect is that a pure text master model will be displayed in the model selection machine to support the picture (this is intentional so that the message can be released).
Install command
dsh plugin --profile web add github:Leeminjing/dsh-eyes