It'll be like this for a lot of AI games as well, and I personally think it'll be more prevalent of an issue with AI devs. It's already the case for many here on F95 too.
These are typically devs with much less coding or gamedev experience. AI tools lower the barrier for entry, no doubt! That's a great thing in my opinion, but these tools will only get you so far. If you rely on them for everything then at a certain point you'll hit a roadblock where it all starts unraveling. Anyone who's dabbled in vibe-coding will have encountered something like this:
Then it fails to compile again, either with the same error you were initially trying to solve, or a new one. Or even worse, it will compile with no errors at all but the code simply doesn't work and does nothing. So you add debug functions to try and verify things during runtime, but those fail too. Lather, rinse, repeat...
--------
Image generation is the most reliable/easiest part of this process, even for the high quality images that this game has. Getting your workflow set up is the toughest part, but once you've got it in place you can crank out any images you want, and so long as you follow the same steps as before you'll achieve the same quality of results every time (for the most part). This is especially true if you're running everything locally, since you'll be insulated from any changes or updates to the online models.
Coding itself is the true challenge and also the main filter here, even for something relatively simple like Ren'py. Things can fall apart very rapidly if you don't have coding experience and you're relying on tools like Claude (or god forbid, ChatGPT lol) to do the work for you. The bigger and more complex your project becomes, the faster its threads will unravel.
There are already a ton of half-baked vibe-coded games here on F95, many of which haven't even lasted more than 2-3 updates before the dev disappears and the project is abandoned. Hopefully Lockstar isn't one of those, but time will tell...
Anyways, good luck, dev! Your image workflows are clearly fine tuned and the results are already awesome and better than the vast majority of GenAI pics I've seen, so if I were you I'd be putting my entire focus on learning the code and ensuring it works for every build. Test, troubleshoot, debug. It will save you a lot of grief from pissed off or annoyed players (like the ones who enter a custom name and are instead told "
Your name is Alex")