--------------------------------------------Divider-----------------------------------------

24.1.25:

1.2.3 is an adjustment based on 1.2.1 (not a fusion based on 1.2.2, 1.2.2 is a standalone version), the biggest improvement is the better sensitivity of prompting words, to a large extent, avoiding the awkwardness of "not drawing the intended meaning", sometimes if you're not satisfied, just draw more cards.

In addition, according to the webui testing deployed locally, Lora's compatibility is better than 1.2.1, but I didn't use the online map test of lib. I hope everyone can provide more feedback to help with testing.

The built-in VAE is vae-ft-ema-560000-ema-pruned. It seems there's not much difference compared to 840000.

As for the effect of the generated images, it's a minor adjustment. Frankly speaking, with the current effect, if you want to make significant progress, you may have to train the images on this basis or make significant changes to the drawing style. Not just my model, but all fusion models are facing bottlenecks. Mixing and matching, you will find that the differences between models of the same type are getting smaller and smaller. Running a few images locally, no matter how you look at it, it feels like d￩j￠ vu...

Please ignore the bad hands of a few example images, no time to fix them.

In the future, I will slowly release a collection of natural outdoor photo shoots of girls generated with version 1.2.3.

I pursue a sense of realism, with colors as vivid as possible, skin as transparent as possible, and as good compatibility with Lora as possible. For friends with suggestions or opinions, feel free to leave a message for communication.

The models and image collections I publish are purely for my own entertainment...