Make sure to use UI that supports v-prediction, otherwise you will get a noisy blob!See about version for credits!<h3 id="main-parameters:">Main parameters:</h3><ul><li>Steps: 28</li><li>CFG: 4</li><li>Sampler: euler a (euler for v6 and lower)</li><li>Scheduler: normal</li><li>Im also using CFGZeroStar and sometimes Self-Attention Guidance: default on difficult backgrounds)</li></ul><h3 id="positive-prompt:">Positive prompt:</h3><pre><code>number of characters (1girl/1boy), character name, artist prompt, scene/environment/camera angle, action, expression, items, quality prompts (masterpiece, best quality)</code></pre><ul><li>some heavily stylized characters might require prompt: realistic</li><li>you can also try to put nose into faceDetailer</li></ul><h3 id="negative-prompt:">Negative prompt:</h3><pre><code>lowres, worst quality</code></pre><ul><li>url against watermark</li><li>lineart, sketch, dot nose for realism</li></ul>Enjoy..

[BPM] Blink's paradise merge (v-pred)

Corrected dataset, tried higher dim and gradient accumulation 2 instead of 1 for effective batch size of 8 instead of 4. Better anatomy vs v11. Models used, go like them!<ol><li><a target="_blank" rel="ugc" href="https://huggingface.co/Doctor-Shotgun/NoobAI-XL-Merges/blob/main/noobaiXLNAIXL_vPred10Version-cyberrealistic4-perpendicular.safetensors">Doctor-Shotgun/NoobAI-XL-Merges</a></li><li><a target="_blank" rel="ugc" href="https://civitai.com/models/520661?modelVersionId=1979724">2DN</a></li><li>My LoRAs</li></ol>

Files with SHA256: 6c1630d0458b3e8b0f37056696d4a310d902f29c020271bca77fcd20796dc112