We are looking to automate our UI/UX testing using generative AI. Can Gemini accurately identify visual regressions or non-compliance with our design systems? I am particularly interested in how it handles "edge cases" that traditional automated scripts often miss. Could this replace our manual QA team, or should we be looking at a hybrid approach where AI does the heavy lifting for smoke tests?
3 answers
Gemini’s vision capabilities make it excellent for visual QA. Traditional scripts look for ID tags in the code, but Gemini looks at the screen like a human does. It can detect if a button is overlapping text or if the color contrast doesn't meet accessibility standards (WCAG). However, it shouldn't replace your manual QA team entirely. Use Gemini to run 24/7 "exploratory" tests to find obvious visual breaks. Your human analysts should then focus on the user experience—the "feel" of the app—and complex logic paths that require a deep understanding of the end-user's intent and emotional response.
Are you worried about the cost of running vision-based tokens for every single test build, or have you found a way to optimize that?
We use Gemini to generate the test cases themselves. We feed it the user stories, and it writes the Gherkin scripts for us automatically.
That's a great use case, Thomas. It ensures that the tests are always aligned with the latest requirements without a BA having to write them manually.
Christopher, that is a major concern. To optimize, we only trigger the "Gemini Vision" tests for major UI changes or once a day for a full regression sweep. For minor logic-only builds, we stick to our standard Selenium scripts. This hybrid approach keeps our cloud costs manageable while still giving us the "visual safety net" that generative AI provides. It’s all about finding that balance between the high cost of sophisticated AI tokens and the manual labor cost of a human tester.