This model was trained using extremely low resolution screenshots of the show "Avatar The Last Airbender" as I was getting some pretty severe artifacts when attempting to upscale. I'd reccomend a weight of 1 generally I've found that the range .9 (To fix character generations) and 1.15 (To force style) tend to work alright. <h1 id="v1-notes">V1 Notes </h1>This lora was not trained using any of the names of the character's within the show and therefore will not know the characters implicitly.<h2 id="strengths-and-weaknesses">Strengths and weaknesses</h2>Strengths:<ul><li>Decent quality environment and background</li></ul>Weaknesses:<ul><li>Strongly conforms to any stylistic tags in the prompt. (If you prompt for photorealism, you'll get photorealism)</li></ul><ul><li>Struggles with characters in a variety of ways.</li><li>Tries to render the characters in the same high detail manner of the environments in close-up shots</li><li>Completely distorts the characters from a distance</li><li>Tends to close the eyes of the characters without being prompted to do so.</li></ul><h2 id="prompt-notes">Prompt notes</h2> If you'd like to see some bending you can use the following prompt scheme:<ul><li> {subject} casts fire from their hands.</li><li>{subject} is casting a spell causing water to swirl around them violently.</li><li>{subject} is casting a spell that causes the boulder to levitate.</li><li>etc.</li></ul>

Files with SHA256: 1c59b76e0863379f52c40c2409fdcfd9383f6441eef0ec22235b1d56547c0612