This LoRA was trained based on Flux-1-dev. Due to issues with the dataset quality and some underfitting, my initial assessment is that it might be caused by insufficient training steps and learning rate. This LoRA was trained based on Flux-1-dev. Due to issues with the dataset quality and some underfitting, my initial assessment is that it might be caused by insufficient training steps and learning rate. Flux LoRA models are generally less prone to collapsing during training. However, it does seem quite sensitive to the quality of the dataset.The blurry images are likely due to the dataset quality—currently looking into it. Would appreciate any experienced insightsFlux LoRA models are generally less prone to collapsing during training. However, it does seem quite sensitive to the quality of the dataset.Inspired by 《BIT》<h1 id="v1">V1 </h1><h2 id="recommended-weight:-1">Recommended weight: 1</h2><h2 id="recommended-settings">Recommended settings</h2>Sampler: EulerSteps: 20CFG: 1Trigger words：ww1pon<h3 id="prompts">Prompts</h3>When writing your prompt, put the trigger word at the very beginning and use natural language for the rest.For example：<pre><code>ww1pon, a side view of a compact sci-fi ray gun, silver and light blue body with cyan line accents, large spherical blue glass dome on top with a circular targeting lens, short barrel with a small muzzle, angular trigger area, light gray grip with dark pad, small cyan blade attachment underneath, white background</code></pre><pre><code>ww1pon, a side view of an energy lance weapon, black and yellow color scheme, sharp golden trident-shaped energy blade at the tip emitting yellow glow, black segmented shaft with yellow accent details, long dark handle, white background</code></pre>

Files with SHA256: 1402f976936604c41f6d0a840fddc30c176ba114827e980e137c5ad0b890198a