Images in conversation are too compressed for pixel-level CSS matching tasks

Problem

When a user shares a screenshot to describe a desired UI design, the image rendered in the conversation context is compressed/downscaled to the point where fine visual details are unreadable:

• Border thickness and style (solid vs dashed vs dotted)
• Opacity values
• Letter spacing
• Subtle background fills

This causes Claude to loop on CSS tweaks indefinitely, making wrong guesses each iteration, instead of detecting early that the image resolution is insufficient.

Expected behavior

Claude should detect when an image is too low-resolution for pixel-level design work and proactively ask for a zoomed screenshot of the specific area before attempting any CSS changes.

Alternatively, images shared in conversation should preserve enough resolution for design reference tasks.

Impact

In practice this led to 10+ failed CSS iterations on a simple border/background styling task that should have taken 1-2 attempts.