|
>>
|
No. 1517
File
175884298976.png
- (1.10MB
, 984x695
, mascollage real fix.png
)
>>1515
Yeah. It all seems to involve making the right prompt - as someone said in a site I frequent (speaking about ChatGPT as a tool), it's like doing a commission to an Indian techie. You need to guide LLMs carefully so you get the desired results.
Ultimately, it's all about being satisfied with the results. The pic to the left, for example, only had AI do the face-swapping. The rest was just very basic (like, MS Paint basic) photomanipulation and organizing into a collage to create the illusion of a spy preparing to wear a mask of someone else. It's not perfect but it's enough to suspend disbelief and tell a story. With better skills, the scene would've developed a lot better. With a good prompt, AI could've probably done this (the three pictures are mine, using a larger mask as the prop and actually wearing it at the end), but it would've taken a different amount of work and patience, probably for the same results.
Sure, LLMs and photomanipulation/image creation AI tools will most likely get better when fed all that data, but ultimately it'll stop at a point where the majority of people feel satisfied.
|