Screenshot Describer

A vision model doesn't take a file path. It takes bytes, turned into text, sitting inside the same message list you already use for text-only prompts. This lesson builds the one new piece every later lesson depends on: getting an image into the request correct

8 lessons, each with runnable code in the browser.

  1. Sending a Screenshot to the Model
  2. Reading the Reply's Content Blocks
  3. Prompting for a Structured Description
  4. From Description to Starter Markup
  5. Turning Regions into Layout CSS
  6. Harden: Fixing Broken Descriptions
  7. Harden: Big Screenshots Cost Tokens
  8. Ship the Screenshot Describer

Compilearn home