SD 1.5 trained with dreambooth on dataset made up of this images https://www.instagram.com/a_thousand_d__s

It's a collaborative project where everyone can modify communal vision by submitting their own images.

Use 'ohwx' token to generate something that has a ??????? shape. Keep in mind that training set had a broad range of such things and, therefore, doesn't converge to anything specific (people see things differently - and that's the main point of this affair), thus some generated images may be quite far from commonly accepted ??????? shape or plainly irrelevant to you.

DPM++ SDE Karras sampler works best to my opinion.