This model is focused on replicating Dagasi's style with maximum flexibility, but it also recognizes their OCs with good accuracy. Use e621 tags to prompt OCs. Please publish generated photos or leave a review about your experience with my model; I don't use this site much but I enjoy ai training and wanna learn how I can improve 👍<h3 id="updates:-ov2y0nx25">Updates:</h3>v1: initial release.v2: Increased lora size, increased training steps, and lowered learning rate. Appears to capture fine details better with improved anatomy.v3: Increased lora size, increased training steps, and increased learning rate. This version seems to have better flexibility and more consistent style, especially around the eyes. This version, however, has also captured the concept of spiral eyes, and tends to partially apply it to the OC characters. If this poses a problem during generation, try adding "[@_@]" to the negative prompt.v4: Lowest learning rate so far, slightly tweaked noise and min_snr, otherwise the same. This version exhibits the best flexibility and anatomy so far.v5: Not happening. While messing with the dataset I lost the original, and I'm not going to repeat the infuriating task of curating tags again. If you have any requests or questions about the lora, feel free to message me.

NoobAI Dagasi style

Files with SHA256: 02d1b4e4f533883970e5e3553743d041e24ebc688b567d52a34551fe18e11528