This episode is a hands-on evaluation of OpenAI’s GPT-Image-1.5, with a running comparison to Nano Banana Pro using lots of small, practical tests.
- Prompt-following and “knowledge” checks (including character/subject familiarity).
- Filter and refusal behavior.
- Spatial reasoning and multi-object composition.
- “Utility” tasks like screenshots/interfaces, diagram generation, table → chart, and basic coding-related comprehension.
- A few visual search / puzzle-style prompts (e.g., Where’s Waldo).
If you’re trying to decide whether a new image model is meaningfully better for day-to-day work, the value here is the breadth of test cases: it’s less about a single benchmark score and more about where failures still show up (text, layout, consistency) and which workflows feel ready to use.