Field notes from people using GPT Image 2 for real work — prompts, comparisons, and the things we wish we had known earlier.
A week of testing GPT Image 2 against the work I actually do every day: thumbnails, ad creatives, and a couple of book covers. Where it nailed it, and where I had to fight it.
Specific phrasing patterns that consistently improve GPT Image 2 results — written after a couple of hundred generations and a lot of bad ones.
An honest comparison after using GPT Image 2 alongside Midjourney, DALL-E 3, and Stable Diffusion for a working month. No tier lists. Just notes.