All projects

Workshop pilots

Visual AI Evaluation

Reference-led demonstrations compare AI imagery, identify factual or visual errors, and turn human critique into specific corrections.

Inside the project

Look closely enough to catch what the model invented.

  1. Establish the reference
  2. Inspect the output
  3. Describe the mismatch
  4. Review the correction

Where it started

An image can look polished while getting the product, label or physical detail wrong. These demonstrations bring photographic judgment to that gap.

What took shape

The workflow compares an authoritative reference with a generated result, names the mismatch and directs a narrower correction. Packaging and product details make the difference between a plausible image and a faithful one visible.

What the work revealed

Reference authority matters more than confidence in the output. A correction needs another comparison, because repairing one area can introduce an error somewhere else.

The next step

Extend the demonstrations with examples whose references and publication permissions are clear.