Long story short, to make it work on ChatGPT, user need to ask the correct question after uploaded a photo.
To put into context, I am using ChatGPT free account that will require sign in, but I do not pay for it.
A correct question needs to be asked in order for it to happen.
Here is the magic statement to ask are as follows:
The first generated photo from observation is the most 70% accurate, background is mostly correct but with an incorrect flag. There is no Indonesian flag in original photo. On top of, it instead of 9 persons, only 7 were added. Out of 7 faces only one is correct. Clothes are incorrect.
Using prompt to rectify, here are the 2nd attempt here are the result.
It is still 70% accurate, background is better but added ghostly partial face is added to the background. There is not so much of flags in the original photo. I guess due to my prompting ChatGPT decided to go full right-wing boner. Most of the folks are barely recognizable not even a correct caricature. Clothes are still mostly wrong.
Even my friends without the reference to the original photo will not able to identify me.
The observation and hypothesis to make the Ghibli styled to be more accurate are but not limited to:
- Provide photo will less persons, one is the best, two are ok.
- Higher resolution photo in order for ChatGPT to able to extract face and the person facial feature in the photo.
Unfortunately, for free account, only 2 image generation per day or 2 images per 24 hours.