<h1 id="overview">Overview</h1>The goal of StoryMix is where the boundaries between fantasy and reality merge. With a humble start by discovering its current capabilities and flaws. Future updates will be focusing on improve the model and make it easier to use. Positive Prompt: <code>best quality, masterpiece, ultra high res, (photorealistic:1.4), raw photo, sharp focus, HDR, detailed skin</code>Negative Prompt: <code>Anime, cartoon, ng_deepnegative_v1_75t, (badhandv4:1.2), (worst quality:2), (low quality:2), (normal quality:2), lowres, bad anatomy, bad hands, normal quality, (monochrome), (grayscale) watermark.</code> Merge Based On:<ol><li><a target="_blank" rel="ugc" href="https://civarchive.com/models/8124/a-zovya-rpg-artist-tools">https://civarchive.com/models/8124/a-zovya-rpg-artist-tools</a></li><li><a target="_blank" rel="ugc" href="https://civarchive.com/models/47130?modelVersionId=70157">https://civarchive.com/models/47130?modelVersionId=70157</a></li><li><a target="_blank" rel="ugc" href="https://civarchive.com/models/43331/majicmix-realistic">https://civarchive.com/models/43331/majicmix-realistic</a></li></ol>(Optional) Recommended VAE to get a better high quality result, please use this VAE:<ul><li><a target="_blank" rel="ugc" href="https://civarchive.com/models/76118/vae-ft-mse-840000-ema-pruned">https://civarchive.com/models/76118/vae-ft-mse-840000-ema-pruned</a></li></ul>(Optional) Recommended Textual-inversionto prevent bad hands and ugly results please use these textual-inversions:<ul><li><a target="_blank" rel="ugc" href="https://civarchive.com/models/16993/badhandv4-animeillustdiffusion">https://civarchive.com/models/16993/badhandv4-animeillustdiffusion</a></li><li><a target="_blank" rel="ugc" href="https://civarchive.com/models/4629/deep-negative-v1x">https://civarchive.com/models/4629/deep-negative-v1x</a></li></ul>Steps used when I generate images using this model<ol><li>Start prompt engineering<ol><li>Using 'txt-2-img', or 'img-2-img' using existing images &amp; prompts if you want</li><li>Start simple without <a target="_blank" rel="ugc" href="https://github.com/Bing-su/adetailer">After Detailer Extension</a></li><li>Settings:<ol><li>Sampling method = Euler a / DPM++2MKarras</li><li>Sampling steps = 20-60</li><li>CFG = 5-7 for more creativity, 10-15 to be strict</li><li>Size: 920 x 512 px (portrait) - Or 512x920 landscape mode that tend to generate more characters - Or 512x512 if you prefer to use the result for training</li><li>Start with batch count = 1-4</li></ol></li><li>Edit the prompts until reaching the intended results<ul><li>(Recommended) Ask chatGPT to generate text-to-image prompts suitable for Stable Diffusion according your requirements</li></ul></li></ol></li><li>Polish the Details (if needed)<ol><li>Keep using 'txt-2-img' with the config in step 1<ul><li>or choose the image you like and 'Send to ImgToImg' if you want it to be very similar</li></ul></li><li>Reuse seed from the previous image (Useful when we want to keep some major details like faces)</li><li>Use the <a target="_blank" rel="ugc" href="https://github.com/Bing-su/adetailer">After Detailer Extension</a><ol><li>ADetailer Model = yolov_8n<ol><li>For most cases, first detailer is for face</li><li>Second detailer is for hands</li><li>Set Confidence = above 65 to avoid false detection</li></ol></li></ol></li><li>Settings:<ol><li>Sampling method = Euler a / DPM++2MKarras</li><li>Sampling steps = 40-100 for a better quality</li></ol></li><li>(optional) Use DeepBoru to improve prompts but be careful to also check missing prompts not covered by DeepBoru</li><li>(optional) Fix minor flaws with inpainting/sketch/control-net or other tools<ol><li>Inpainting:<ol><li>Erase area to improve</li><li>Then generate with high resolution</li></ol></li><li>Sketch/Control-net: rarely used, but can be useful to improve minor flaws:<ol><li>Bad hands can be fixed with control-net depth map</li><li>Bad pose with control-net OpenPose</li><li>When prompt is hard, minor tweaks can be improved with sketch, e.g. adding missing legs or characters</li></ol></li></ol></li></ol></li><li>Send to Extras<ol><li>Scale to = 1080 x 1920 / 1920x1080</li><li>Upscaler1 = SwinIR_4x or other 4X scaler</li><li>GFPGAN visibility = 0.95 </li><li>(If it needs cropping) Crop to fit = checked</li></ol></li></ol>

Files with SHA256: e33ca146ff37addce45332ff1db61d379fb1cd3d2660fb92ec1c6411fc9e42f4