This LoRA was born from the desire to recreate a dress I saw some time ago on a model: a garment made of thin horizontal strips, held equidistant by vertical bands, that shaped themselves to the body of whoever wore it.

To build it, I first put together the most varied dataset I could from the few images I had available. Working on the captions, I tried to isolate the concept of the striped dress from everything else present in the photos. The first version, however, had also learned the physical features of the original model.

So I used this v1 to generate a new, much more heterogeneous set of images, which I then used to retrain the model. The result is that it went from a single initial dress to a potentially infinite range of garments that keep the same style. And not only that: the LoRA can also be applied to stylize objects, buildings and other subjects, going well beyond clothing.

How to use it

Trigger word: l4dderdress

• Recommended strength: 1.0 (I haven't experimented with higher values)

• Basic prompt: l4dderdress, [subject],

Try it on different subjects and experiment: it shines when you push it outside the context it was born for.