Generates ankle socks reliably. Current models apparently only generate calf-, knee- or even thigh-high socks when prompted with "ankle socks" or "ankle_socks."
Trigger words are "ankle socks" or "ankle_socks." This version uses the same core dataset as v1.0, but with attention masking to improve the training process and subsequently, model flexibility. Model was trained in OneTrainer using the DADAPT-LION scheduler with a constant LR of 1.0 over a dataset of 168 images with a batch size of 12 and a gradient accumulation of 16 steps for a total of 290 epochs (4060 steps).
LoRa weights were normalized at 1.0 (the recommended weight to use this model) using the invaluable Stable Diffusion ai-toolkit (https://github.com/ostris/ai-toolkit).
Sample images include all metadata for recreation in your own environments; highres. fix and ADetailer (just to make faces pretty) were used to generate the samples. No other embeddings, LoRa or manual inpainting were used during/after generation.
You're free to use this however you wish. It's socks. Would love to see some creations using this LoRa!