1 DeepSeek is powerful, but has one drawback
DeepSeek is really impressive this time. It is affordable, and the official release of V4-Flash brings another major performance boost. According to the OpenCode team, DeepSeek V4-Flash reached a daily token consumption of 8 trillion on its platform, setting a new single-day record.
But it has one drawback: it is a text-only model and cannot recognize images.
However, image recognition is important in programming development, bug fixing, UI recognition, and other scenarios. In these situations, handling things through text descriptions alone is not very convenient. If it can understand screenshots directly, the process becomes much faster and more convenient, greatly improving efficiency.
2 Open-source image recognition Skill: Claude Vision
Here is an open-source image recognition Skill—claude-vision-skill. Although it was originally designed for Claude, that does not matter: it can be installed in any Agent, which will automatically adapt to it.
https://github.com/asuojun/claude-vision-skill
— Skill repository page
This Skill mainly gives image recognition capabilities to models that do not have them.
How it works: the image is sent to a model with image recognition capabilities, which returns a textual description. DeepSeek then processes that description, so an image recognition model must be configured.
The built-in recommendation is Qwen (qwen3.5-omni-plus, qwen-vl-max) through Alibaba Cloud Bailian. New users receive 1 million free tokens, costing approximately 0.02 yuan per use.
— Built-in recommended Qwen models
You can also use other image recognition models that support the OpenAI-compatible format.
3 How to install and configure it
How do you install it? It is simple: just give the repository URL to Codex, and it will install and configure it for you.
— Give the repository URL to Codex for automatic installation
By default, it uses Alibaba Qwen’s image recognition model (qwen3.5-omni-plus). New users get 1 million tokens, which is enough for approximately 7000 image recognition requests.
To get an Alibaba Qwen API Key, open the Alibaba Qwen console, create an API Key, and copy and enter it.
https://platform.qianwenai.com/home/api-keys
— Create an API Key in the Alibaba Qwen console
4 Installation complete: testing the results
Once installation is complete, you can use it directly.
— Ready to use after successful installation
Let’s test it by asking it to recognize the content in this image.
— Image used for recognition testing
The result is shown below, and the recognized content matches completely.
— Recognition result matches completely
DeepSeek now has “eyes” and can recognize images, taking its efficiency to the next level.
Conclusion
By adding image recognition capabilities to DeepSeek through an open-source Skill, you get immediate results at an extremely low cost. From now on, programming, debugging, and UI recognition can all work directly with images, allowing a text-only model to truly “see” the world.