<h1 id="dragon-ball-super-qwen-image">Dragon Ball Super - Qwen Image</h1>This was my first attempt at training a LoRA for Qwen Image. The results are promising for the overall anime style, but character faces perform average.<h2 id="trigger-word:-dragon_ball_super">Trigger Word: dragon_ball_super</h2><hr /><h2 id="coverage">Coverage</h2><ul><li>Almost all Dragon Ball Super (DBS) characters were included.</li><li>Not all have been tested yet.</li></ul><hr /><h2 id="works-well-when">Works Well When</h2><ul><li>Resolution: Use <code>1280x720</code> landscape (dataset was trained only on this resolution).</li><li>Prompts: Describe character by name and appearance.<ul><li>Example: Bulma, a woman with short blue hair</li></ul></li><li>Character Count: Solo characters or maximum 2.</li><li>LoRA Strength: <code>0.8 – 0.9</code></li></ul><hr /><h2 id="gotchas">Gotchas</h2><ul><li>Goku and Vegeta often get mixed up → use negative prompts.</li><li>Always describe famous characters’ look and outfit explicitly.</li></ul>Other Versions: <a target="_blank" rel="ugc" href="https://huggingface.co/ksol8/dragon_ball_super_qwen_image/tree/main">https://huggingface.co/ksol8/dragon_ball_super_qwen_image/tree/main</a>

Files with SHA256: e08aa6f1530e69c22923f2885ef953186a479fa7c15f20f2ec05749544809eaf