ChatGPT's image generation gets searched almost as often as Nano Banana's for a simple reason: a huge number of people already have ChatGPT open for something else and try the photo prompt there first, before ever looking for a separate tool. That habit alone makes it worth understanding on its own terms rather than treating every AI photo prompt as interchangeable across platforms.
The most useful thing to know about ChatGPT specifically is that it behaves conversationally. Rather than treating each prompt as an isolated instruction, it treats the conversation as an ongoing project you're refining together. That makes it well suited to a back-and-forth workflow: generate a first version, then describe exactly what to adjust in the next message instead of rewriting the whole prompt.
ChatGPT also tends to handle legible text inside an image more reliably than most alternatives, which matters for anything poster-like — a title, a birthday message, a name on a jersey. If your concept depends on readable text appearing correctly in the final image, that's a point in ChatGPT's favor over tools that are stronger at pure photorealism but weaker at typography.
For a first-message prompt, the same core structure that works elsewhere still applies: describe the subject and identity requirement, the transformation or scene, the composition, and the lighting or mood. What changes is how much you lean on the follow-up messages afterward. Where a single-shot tool needs everything specified up front, ChatGPT rewards leaving room to adjust after seeing the first result.
Identity preservation works a little differently here too. Rather than a single dense instruction, it can help to state it plainly and then reinforce it if a follow-up edit starts drifting — for example, if you ask for a background change and the face shifts slightly as a side effect, a short follow-up like “keep the face exactly as it was in the previous image, only change the background” is usually enough to correct it.
One common frustration people run into is over-stylization: asking for a specific aesthetic and getting a result that leans further into that style than intended, at the cost of realism. If that happens, it's more effective to explicitly ask for the realism to be dialed back in a follow-up (“make this look like a real photograph, reduce the illustrated or painted quality”) than to rewrite the entire original prompt.
A ChatGPT-specific prompt formula
Open with subject, identity requirement and the core transformation in one message. Keep the first prompt moderately detailed rather than exhaustive — composition, lighting direction and mood are usually enough. Save fine-tuning (color grading, minor pose adjustments, background swaps) for short follow-up messages rather than trying to lock everything in on the first attempt.
Getting good iterative edits
Be specific about what should change and explicitly say what should stay the same in the same message — “change only the lighting to golden hour, keep the pose and outfit exactly as they are” works better than “make it more golden hour.” Vague follow-ups tend to produce broader, less predictable changes than the original prompt.
When to reach for ChatGPT versus another tool
ChatGPT is a strong choice when a project involves several rounds of back-and-forth refinement, when text needs to render correctly inside the image, or when you're already mid-conversation about the concept and don't want to switch tools. For a single, carefully pre-planned identity-preserving portrait where you already know exactly what you want in one shot, a tool built specifically around photo editing can sometimes get there in fewer attempts.
CreateLoom's prompt library flags which tools each prompt was written for so you can pick based on how you actually want to work — one detailed attempt, or a conversation you refine as you go.
Ready to create something?
Browse detailed AI image prompts and find a starting point for your next creative idea.
Explore AI Prompts →