>>3019163>be me>spend years trying to improve my lora baking in order to properly capture a style>finally feel I've reached the peak and I'm merely playing a game of model swapping>remember some anon a while back singing the praises of fine tuning over loras>finally decided to give it a try using the Seele model>had no idea what I was doing so I just used similar settings to lora training>first results were mediocre at best, but I saw potential>after tinkering with settings and image count, I realized using more images is better>got to the 5th iteration and the results were mind blowing>not only did the model perfectly capture the style, the model also stabilized, and the backgrounds improved dramatically>as a last little experiment, I retained some loras using the same data from the fine tune and applied it on top of the fine tune resulting in further improvementThis hobby never fails to excite me.
