Question about Resolution Support Implementation

[Han1018]

Hi,

Thank you for your amazing work on this project!

I have a question regarding how the project supports different screen resolutions. From what I understand by tracing the code, it seems like the screen information is provided directly to Claude without additional compression or pre-processing. Could you kindly confirm if this is the intended approach? Or is there any additional mechanism involved that I might have missed?

I’m asking because I’m trying to ensure I fully understand the implementation and its scalability across various resolutions.

Looking forward to your response🙏

[h-siyuan]

The screenshot mechanism differs by model. For example, Claude Computer Use by Anthropic is extremely costly, thus to help users save token cost, we resize the screenshots to low-resolution ones. For Local Run with ShowUI, as the token cost is far cheaper, we keep the original screenshot sizes for better performance:)