To prevent further prompting, I tracked the monitors and composited in a clean, accurate developer dashboard. Same actor, same shot, real software, with the ability to edit and iterate the screens to their exact specifications. Real iteration, not guesswork. Adjust the UI as many times as you need — no re-generating, no rolling the dice.
Drag the slider to see before/after
The client was happy with the AI-generated talent — the look, the performance, the setting — but the screens behind him were a mess. Misspelled headers, charts with percentages that don't add up, and dates that don't exist. Classic AI-generated UI: convincing at a glance, falling apart under any real attention.
Re-running the generation to fix it would've meant more render costs and more time waiting on the prompt lottery, with no guarantee the next pass would look any better.