Course contents
Images at Work
GPT Image 2 for real image jobs: targeted edits, queued changes mid-generation, and pinned comments.
Work's image engine (GPT Image 2) is built for jobs, not just generation: hand it an existing image and change only what you name. "Change the text to the new title, swap the logo for this one — only change those things, give me three variations" preserves everything else. That only-change-X discipline is the whole trick for thumbnails, banners, product shots, and social graphics.
Two workflow features beat re-prompting. While a generation runs, you can queue further edits — hit edit and add "brown shirt on the left, reduce the glow" without waiting. And on a finished image, pinned comments give spatial feedback: click the exact spot, leave a note there ("darken this corner", "whiten teeth"), and the revision applies feedback where you pinned it — the design-review workflow, minus the designer round-trip.
Images compose with everything else: generate the variations, then "email all four to Emily for a vote" via the Gmail plugin, or drop the winner straight into the deck the agent built an hour ago.