A New OpenCode Plugin Gives Text-Only LLMs Vision by Routing Images to a Multimodal Model

I've been using OpenCode for a long time, and whenever I use open-source models like DeepSeek or GLM, there's always a deep sense of powerlessness. Although these models are very strong, they are text-only, which directly prevents the AI from solving problems that involve image recognition scenarios.

So I casually developed a multimodal plugin for OpenCode. It adds a tool to the current chat context. When the model recognizes that it needs to identify an image and the current model is on the list of models that need image recognition (which you can configure yourself), it will call this tool to invoke a model capable of image recognition and return the result into the context body. This indirectly achieves the image recognition effect. This plugin is already open source, but it hasn't passed the official ecosystem verification yet. For now, you can only download the source code and drop it into a specific OpenCode folder to enable it.

Talking so much is useless, let me just serve it up: https://github.com/cipherTing/opencode-analyze-image

If you encounter problems, it's recommended to just post an issue on GitHub. If you have any questions or need to discuss something, you can communicate in the comments section.