Conceptual Photography

V9Like the original Ghost, it is a bit overtrained, but its good to lower the strength and mix it with other loras. The dim is also big, and like the first krea2 training the eyes/face can get messed up at full strength (it was trained at 512px which seems to still be the best for some reason). Trained on real, PG-Rated photography. No ai in the training images.After a week of training and testing I have found a good epoch that:-captures my photography style-has decent seed variety (more than vanilla krea2)-is interesting, and gets creative with characters interacting (this was SO LACKING in flux, so happy to have some of this in krea2 as my photography features a lot of multiple characters interacting)-unique compositions

ghost-v2.safetensors

v9-(ghost v2[kr2])

v8-(dirty pastel [kr2])

v7 - (infrared [f1])

v3.1 - (evil robot [f1])

v6 - (ghost [f1])

V2.1 - (angular v2 [f1])

v5.1 (junktopia v2[f1])

v5 (junktopia[f1])

v4 (monsterbeam[f1])

v3 (neon darkness[f1])

V2 (angular[f1])

V1 (love and wolves)

Files with SHA256: d9a190ccf557aeb82b9dffe9252aa63cf6c4b5e78a95b6a4074979b0827fda08