I’ve been spending two days trying to generate an image where a civilian man is walking away down the street. He’s wearing a coat without any belt, a classic cap, and high boots with wide trousers tucked into them.
The result: Around 57 images (I might have miscounted slightly, but not by much) and at least as many prompts (probably even more), yet every single time the character is depicted as a military figure — in a soldier’s or officer’s overcoat, boots, or cap. In other words, no matter how I phrase it, the result is never a civilian. The best outcome I managed to achieve was a civilian, but still with a belt on the coat. It seems like the system is completely stuck on including the belt. I’ve tried rephrasing the prompt multiple times, emphasizing that it’s supposed to be a civilian man, but DALL-E seems to completely ignore this.
To test the issue, I even deliberately specified “military” in the prompt to see if DALL-E might be confusing the terms “civilian” and “military.” But no — when I specify “military,” the system produces a proper military figure. What’s more frustrating is that the system has previously created images of this same character correctly, but even if I use those as a reference, nothing changes — it still ignores them just as successfully.
Honestly, I’m at a loss and don’t know what else to try. I’m stuck.