App Store review — ChatGPT (4★): Great creative potential, exhausting inconsistency
retained text
4★ review of ChatGPT: When ChatGPT works well, it can be a fantastic creative tool. Lately, however, I’ve spent more time correcting recurring problems than making progress on my projects. I maintain a Character Bible with reference cards and established descriptions. Despite that, ChatGPT repeatedly changes my characters’ faces, body types, hair, eyes, outfits, and identifying details. Sometimes it effectively substitutes a different character. Asking for an art-style update becomes an unwanted redesign, even when I explicitly say to preserve the original face and body. Memory problems make this worse. Incorrect information has resurfaced after I corrected it and deleted the bad memory. A character with purple-and-golden-blonde hair was described as a redhead again, and that mistake made its way into lyrics written for Suno. These failures undermine images, songs, stories, and the continuity of an entire project. Model selection is another frustration. Despite selecting ChatGPT 6, new chats keep opening on 5.6 in my experience. I shouldn’t have to repeatedly check whether the app is using the model I selected. If that choice does not carry over, the interface should make that clear. The handling of mature content also feels like a moving target. A request involving fictional adults may be accepted one day, while a comparable request triggers repeated “too sexual” refusals the next. I’m talking about material I understand to be mature but non-explicit, not a demand for unrestricted content. I cannot tell whether the rules are changing or their enforcement is inconsistent, but the practical result is the same: confusing boundaries and wasted attempts. Even after working with the assistant to clarify a request, I can still encounter another block. Adult users deserve clear explanations and predictable enforcement. Meanwhile, seeing OpenAI continue posting promotional YouTube Shorts is frustrating when I still don’t have a usable replacement for Sora. Promotional activity is not a substitute for the creative capabilities users are waiting for. I would rather see clear communication about availability and limitations than more marketing while gaps in my workflow remain unresolved. The recurring cycle is exhausting: explain the mistake, provide the reference, receive an acknowledgment, try again, and encounter the same problem—or discover that fixing one detail broke something that was already correct. I’ve also dealt with duplicate images, missed poses, and obvious anatomy errors. The gap between reassurance and results is especially frustrating. Being told a correction is understood means little when the next output contradicts it. Users need accurate explanations of what was actually checked or saved, and what the system cannot change. Repeated apologies do not repair recurring failures. The cumulative frustration has become bad enough that I feel I need a day or so away from the app for my mental well-being. This was supposed to be an enjoyable creative outlet. Instead, troubleshooting has left me drained and worn down my enthusiasm for projects I care about. I understand that generative AI is imperfect. I’m not demanding flawless images or finished lyrics on the first attempt. But occasional mistakes are different from repeatedly having to correct the same established facts. Corrections should matter. Rejected information should stay rejected. Model selection and content boundaries should behave predictably. As a paying subscriber, I should not have to act as the assistant’s memory manager, continuity editor, and quality-control department. The creative potential is real, and I’ve received results I liked. That makes the current experience more disappointing: too much of the burden of making this product reliable falls on the user.
document fetched 2026-11-15 05:37