A LORA based on the cover for the album "Porfiado" (by El Cuarteto de Nos). A characteristic concept of a guy riding a horse, backwards, angry as the animal goes in the opposite direction.



I created roughly 15 training images for training, so the LORA is a bit over-baked. Nonetheless, since riding backwards is unusual, reducing the LORA's weight many times causes the riding orientation to be "right" (instead of backwards as in the original cover). This was trained using WAI v12.0 as usual for my LORAs. Anyway, the current version creates valuable results, so I upload it as it is. If I see that the the LORA becomes popular, probably I'll extend the training set to create a version two.

The model has 4 triggers: porfiado_cover, blue_background, white_border, horse. If you replace the "horse" with another animal and increase the animal's weight, you may actually replace the horse; from my experiments, try to add the "riding" tag so that the character actually rides the animal. As commented before, its risky to decrease the LORA's weight, so I recommend 1.0. Overall, the prompting structure should be something like: porfiado_cover, blue_background, white_border, horse, 1girl/1boy, [the character's features]. Take a look at the uploaded examples.

As an additional note, since the character is full body in a reduced section of the image (training resolution 1024x1024), you could benefit from detailers. The results shown here don't use one, but I formatted the prompts with typical quality tags such as "masterpiece,best quality,amazing quality,{prompt}" for the positive and "bad quality,worst quality,worst detail,sketch,censor,{prompt}" for negative.