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. <img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/4bb17a45-bc55-44a9-ac7b-bfb944b06c5c/original=true/4bb17a45-bc55-44a9-ac7b-bfb944b06c5c.jpeg" /> 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 <a rel="ugc" href="https://civarchive.com/models/827184/wai-illustrious-sdxl?modelVersionId=1490781">WAI v12.0</a> 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.

Files with SHA256: f17526b6576745df9f61116b3b7b2587cf6f8f5ef610a4b2d21b387d08c0ffdc