Image Vision (阿里云百炼 Vision Models)

Overview

This skill bundles a Python script (eyes.py) that sends an image to 阿里云百炼 (DashScope)'s vision models and returns a detailed Chinese text description. Use this whenever the user wants help with an image.

Background: The current model (DeepSeek V4 Pro) cannot natively see images. This skill bridges that gap by routing images through Qwen series vision models — the vision model does the seeing, and the text description is fed back into the conversation so the main model can reason about the image content.

When to use

• User sends an image that the current model can't display (shows as [Unsupported Image])
• User asks "what's in this image?", "describe this picture", "analyze this screenshot"
• User provides an image path and wants help understanding it

How to use

Step 1: Run the script

python ~/.claude/skills/deepseek-eyes/eyes.py "<image_path>"

With a custom prompt:

python ~/.claude/skills/deepseek-eyes/eyes.py "<image_path>" --prompt "这张图片里有什么错误信息？"

Switch models or enable high-res:

python ~/.claude/skills/deepseek-eyes/eyes.py "<image_path>" --model "qwen3-vl-plus" --high-res

Step 2: Read the output and answer

The script prints the description to stdout. Read it directly from the command output and answer the user's question about the image based on that description.

Prerequisites

• pip install dashscope (already installed globally)
• DASHSCOPE_API_KEY environment variable must be set
• New 百炼 users get free quota — no payment needed to try

Available Models

Model	Notes
qwen3.6-plus (default)	Latest flagship, best all-around
qwen3.6-flash	Faster & cheaper
qwen3-vl-plus	Dedicated VL, high-precision object recognition

Options

Flag	Description
--model	Switch vision model (see table above)
--prompt	Custom question/prompt for the image
--file-url	Use file:// URL instead of base64
--high-res	High resolution mode (documents/tables/small text)

Notes

• The script always works — the current model never needs to "see" the image directly
• Description prints to stdout, read it directly from the command output
• Default prompt asks for a detailed description in Chinese; override with --prompt for targeted questions
• Switch models with --model to balance speed, cost, and accuracy