Just to be clear right off the bat, I DESPISE this change.
Pros:
- Generated images can be very realistic and of high quality.
- New image size is available.
Cons:
- Generates different people with different facial features every time, there’s no consistency, no matter how many different prompts or entire sentences were provided.
- Even after trying to avoid weird, deformed images one’ll get a few straight-up ugly as sin or absurd images in between a number of generations. I swear I’d never get anything even close to being “ugly” with the previous model, unless I prompted something like ‘Cursed photo’.
- The model can’t draw any resemblance to celeb faces anymore, even if one tries, it generates a random face.
- The amount of time it takes to generate now is around 3x longer, sometimes it just fails and retries.
Lastly, the previous model, which had been in use for over a year, was replaced overnight, like literally. All the prompts that everyone had created over so long and with care, suddenly became meaningless. I had to create an account just to check if it’s only my pain that I’ve to bear.
Me neither. What are you on about?
I gave you proof the seed system preserves details of the image. It is not tied to the style of the image.
It’s a stretched solution and does not count as a positive side of this model either. Actually, you’d know the issue in the first place what I meant by not creating ‘consistent faces’ if you had been working with realistic faces for a year or so, which I doubt
“Stretched solution”? What do you even mean? It is a feature. The old model had a seed system as well, was that a “stretched solution” there too?
And by the way, I was still able to generate similar faces even without using a seed: