<h2 id="recommended-settings-qwen-image-2.1">⚙️ Recommended Settings 🌸 Qwen Image 2.1</h2>Tested with: <code>Qwen Image 2.1 int8_convrot</code><ul><li>LoRA weight: <code>0.8</code></li><li>CLIP strength: <code>0.8</code></li><li>Steps: <code>25</code></li><li>CFG Scale: <code>1.0</code></li><li>Sampler: <code>Euler</code></li><li>Scheduler: <code>Simple</code></li></ul><hr /><h2 id="krea-2">🎨 Krea 2</h2>Tested with: DaSiWa Krea2<ul><li>Steps: <code>10</code></li><li>CFG Scale: <code>1.2</code></li><li>Sampler: <code>Euler</code></li><li>Scheduler: <code>Simple</code></li><li>Denoise: <code>1.0</code></li></ul><hr /><h3 id="illustrious">✨ Illustrious</h3>Tested with: <code>CyberIllustrious Semi</code> and <code>WAI</code><ul><li>LoRA weight: <code>0.8</code></li><li>Steps: <code>40</code></li><li>CFG Scale: <code>3.0–4.0</code></li><li>Sampler: <code>DPM++ 3M SDE GPU</code></li><li>Scheduler: <code>Karras</code></li></ul><hr /><h3 id="anima">🐰 Anima</h3><ul><li>LoRA weight: <code>0.8</code></li><li>Steps: <code>40</code></li><li>CFG Scale: <code>3.0</code></li><li>Sampler: <code>er_sde/normal</code></li><li>Scheduler: <code>Beta</code></li></ul><hr /><h3 id="sdxl-pony">🐴 SDXL / Pony</h3>Tested with: <code>CyberRealistic Pony</code> and Pony checkpoints by <code>Stable_Yogi</code><ul><li>LoRA weight: <code>0.8</code></li><li>Steps: <code>40</code></li><li>CFG Scale: <code>3.0–4.0</code></li><li>Sampler: <code>DPM++ 3M SDE GPU</code></li><li>Scheduler: <code>Karras</code></li></ul><hr />These settings are recommended as a starting point. Adjust the LoRA weight and CFG scale depending on your checkpoint, prompt, and desired style.

Marinette Dupain-Cheng

The training principle has been greatly changed, which has led to improved diversity in image generation

Files with SHA256: 97c900bd95385852ce0eefd8df53f3b426a463f34bbdf9581d3bc61881587d2a