Undressed is a bidirectional SDXL Concept Slider for controlling clothing coverage, from more fully clothed at positive weights to progressively undressed or nude at negative weights.

Unlike a clothing LoRA, Undressed does not impose a particular outfit. It shifts the model's preference for how much clothing the subject wears while generally allowing the prompt and base checkpoint to determine what that clothing is.

Positive weights → more clothed
Negative weights → less clothed / nude

A good starting range is −1 to +1. The current version remains useful considerably farther in either direction; testing has been clean to approximately −3 to +3, although the useful transition depends strongly on the clothing, prompt and checkpoint. In testing, nudity commonly begins somewhere around −0.75 to −2, while intermediate weights can produce partial or intermediate states of dress.

Undressed works especially well when the prompt leaves clothing state ambiguous. It can also push against explicitly described clothing, although strong or concept-defining garments may require more weight. Conversely, it is not intended to override an explicit request for nudity with clothing: explicit prompt semantics can remain stronger than the slider.

Tip — give the slider somewhere to go: for stubborn garments, try wording the subject as “nude beneath [garment]”. This establishes both the body and garment in the prompt's semantic space and can give Undressed a much cleaner path between them. If that wording reveals too much at weight 0, try +0.5 to +1 first to establish the dressed state, then decrease the slider gradually.

You do not have to start at zero. Think of the prompt as establishing the clothing scenario and Undressed as controlling where within that scenario the generation settles.

It has been tested across ordinary clothing, workwear, formalwear, lingerie, layered garments, fantasy clothing, groups, men and women, and a range of poses and compositions. As with other Concept Sliders, the response is semantic rather than a literal frame-by-frame garment-removal animation: garments may open, shorten, disappear, be substituted, or resolve into a different coherent clothing state as the weight changes.

Developed and evaluated on SDXL / Juggernaut XL Ragnarok. Other SDXL checkpoints may respond differently.

Suggested first sweep: +1, +0.5, 0, −0.5, −1, −1.5, −2

Same prompt. Same seed. Move one knob.