If you work with AI coding assistants and frequently pass multiple screenshots, there is a simple way to reduce unnecessary image input.
Instead of sending every screenshot directly to the AI, you can first extract the text locally using Fuji OCR and then send only the relevant text.
The problem
Suppose you have 5–10 screenshots containing:
- Error messages
- Terminal output
- Source code
- Documentation
- Logs
- Configuration files
Sending all those images directly to an AI model can increase the amount of visual input and potentially consume more tokens.
In many cases, you don't actually need the images. You just need the text inside them.
A simpler workflow
You can use Fuji OCR to extract the text locally:
Screenshots → Fuji OCR → Extract text → Send text to AI
Fuji OCR is a Chrome extension that performs OCR locally using WebAssembly.
The OCR processing happens on your device rather than sending the screenshot to an external OCR server.
That means you can:
- Capture the required screen area.
- Extract the text locally with Fuji OCR.
- Review the extracted text.
- Send only the relevant text to your AI assistant.
Why this is useful
This approach can help when you are working with a large number of screenshots.
You can potentially:
- Reduce unnecessary image input
- Reduce token usage
- Extract text faster
- Keep OCR processing local
- Avoid uploading screenshots to an external OCR service
For sensitive development information such as internal logs, error messages, source code, or configuration files, local OCR can also provide an additional privacy advantage.
Example
Instead of:
10 screenshots → AI model
Try:
10 screenshots → Fuji OCR → Text → AI model
The AI receives the information it actually needs rather than processing the entire screenshot collection.
Fuji OCR also supports multiple languages and provides confidence scores for extracted text.
If you regularly work with screenshots while using AI coding tools, this is a simple workflow worth trying.
Top comments (0)