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

24.1.9:

Today's test found that the newly uploaded version 1.2.2 cannot replicate locally generated half-body portrait images using the online tool. The difference between the two is quite significant. Hundreds of images ran locally turned out perfectly, but the online tool even changed the facial shape. The reason for this is still unknown. Let's just use it like this for now.

After all, this version still has its own unique and desirable features. The portrait effect is pretty good in my opinion, and the prominence of abdominal muscles and rib bones has been reduced significantly compared to similar ones, which I am quite satisfied with. Especially when playing locally, it is also very sensitive to NSFW content...

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

24.1.8:

This model is only published on the lib website. If reposted, please indicate the source.

Because my models are all mixed from others' models. In the spirit of open learning and progress, models will be provided for download. They can be freely merged. However, please do not use them for commercial purposes. I do not take any responsibility for the consequences caused by this.

Version 1.2.2 is a fusion of different previous models and different layer parameters. With some slight adjustments to the facial shape and skin tone, this version's upper-body portrait comes with a slight depth of field effect. Some may feel that it is not as good as previous versions; different strokes for different folks. I just played around with merging out of boredom and picked relatively satisfactory results from hundreds of experiments.

As usual, the cover image casually used is a raw image run locally at 512X768, with no facial repair, no high-definition restoration, and no Lora addition.

Local parameters: DPM++2M Karras     Iteration steps: 25     Configuration: 6

Frontal cue words: Canon RF85mm f/1.2, best quality, 1 girl, (Korean mixed, K-pop idol: 1.2), white shiny skin, black eyes, necklace, earrings, (brown long wavy hair), red shiny lips, eyelashes, makeup, shiny, pore, skin texture, bangs, bracelet, off-shoulder, see-through, big breasts

I think at least it's very decent for drawing portraits. Copy the cue words, draw casually, and you hardly need to pick. As for full-body high-definition restoration images, I haven't had the chance to test yet.

Listening to the 4GB VRAM graphics card roaring in agony, it's a pity that I can't train models myself. Fortunately, I'm not a professional; it's just a leisure activity after work...

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