Conceptualized from Michael Costello's 2026 couture design. Trained with generated images and modular caption so you can flexibly modify the color/material of each piece of the gown.Example Prompt:<pre><code>[subject] wearing [overall modification] mcXstrapdiamond gown, with ultra-thin [modifiable] straps over bare torso, connecting to the triangular peak of a [modifiable] skirt exposing hips. Sculptural [modifiable] sleeves wrap around upper arms, tapering down into fitted [modifiable] gloves.</code></pre>Recommended Weight: 0.5 to 0.8Higher weight might influence the general style of the image -- affecting posture, lighting, and tending toward centered composition.<hr />This is a practice for me to train a LoRA almost entirely on synthetic data. Though I tried to make the dataset as diverse as possible, it still seems to have inherited some bias from the base model.

Files with SHA256: 4fe9a8ff12030e597515320f5d819444972614fcc39fa1779f82e7d8e4b81c60