Gemini 3.1 Pro and Opus 4.6: test the brief as well as the model

Keep the three rounds comparable

The 24 February 2026 video tests Claude Opus 4.6 and Gemini 3.1 Pro on a real SaaS landing page. The rounds are a vague redesign request, a stronger designer prompt and a product brief with creative control. The description says weak direction produced little change from either model; it does not supply the full prompt text for copying.

1. Follow the test overview at 0:30 and first results at 1:05 before interpreting the final comparison.
2. Save the same starting page for each attempt and record which model and level of direction are being tested.
3. Define the visitor, product promise and main action so you can evaluate the page beyond how different it looks.

Review design direction and code integration separately

At 4:41 Luke discusses why the prompt is the problem; the second run at 6:10 includes practical issues around lifting sections and merging into production code. A useful design proposal and a safely integrated page are different deliverables. A good-looking section can still conflict with existing navigation, styling or the visitor’s journey.

1. Compare the proposed hierarchy, layout and content with the product brief before moving sections into the application.
2. Inspect the actual merged page and relevant code changes rather than judging an isolated preview.
3. Keep failures or extra integration work beside the visual result when comparing the models.
4. Use the second-round results at 11:32 to identify concrete strengths and weaknesses rather than turn a stylistic preference into a universal ranking.

Use the product-brief round to choose what is worth keeping

The third test begins at 17:00, final comparison at 18:24 and ideas for the real site at 22:45. The source describes one usable result and another weakly styled result, but this written companion does not reconstruct an unseen final page. Review the recording for its exact winning design and the elements Luke would retain.

1. Compare each output with the same product requirements and test the main visitor action.
2. Choose specific elements worth adopting, keeping the rest of the production page stable while evaluating them.
3. Record the accepted design changes and remaining implementation checks before considering the redesign complete.

The takeaway

A stronger brief changes the test. Judge the resulting page and integration effort, not only a model’s reputation for UI design.