This textual inversion was created on the sd1.4 model based on about 100 high resolution images trained in a 512x512 resolution. Images are manually curated photos or video screengrabs. It will work across models but you may need to adjust the weight to achieve the result you are seeking. As weight increases, it will push the output to a more realistic/photo style and then eventually will begin to warp the subject to resemble the original source images embedded in the sd dataset. (Which is distinct and generally low quality.) A setting of 0.7-1.1 should work. If you don't see any effect, increase, if it's too much or looks like trash, decrease. Try a few seeds of course as always. Use "buk1" to invoke.