Text by @sevenof9247 <a target="_blank" rel="ugc" href="https://civarchive.com/articles/3990">https://civarchive.com/articles/3990</a>realvisxlV40_v40Bakedvae - inpainting<h2 id="how-to-use:-9phhbx5ao">How to use:</h2>Recommended Negative Prompt:(worst quality, low quality, illustration, 3d, 2d, painting, cartoons, sketch), open mouthor another negative promptload the modelgo to tab "img2img" -&gt; "inpaint"you have now a view options, i only describe one tab "inpaint"<ol><li>put any image there (below 1024pix or you have much Vram)</li><li>press below "auto detect size" (extention: sd-webui-aspect-ratio-helper)</li><li>draw a maske (you can resize the pencle or reset)</li><li>write a prompt, what you want to see in that mask</li><li>now you have several options</li></ol>a) Masked content<ol><li>fill and latent nothing: best suited for large changes with high denoising strength (&gt;0.8)</li><li>original and latent noise: for minor changes with Denoising (0.4 up to 0.9)</li></ol>b) Inpaint area<ol><li>Whole picture: take more the weight from the whole image</li><li>Only masked: take more weight of your prompt</li></ol>sampling steps ~40, Mask blur 4-8,After multiple rounds of testing, the suggested drawing parameter settings are:<ul><li>Steps ≥ 30</li><li>Sampler: DPM++ 2M Karras</li><li>CFG scale: 7</li><li>Size ≥ 1024x1024</li></ul><h3 id="other-stuff:-fksilbl5r">Other stuff:</h3>Merge betweenrealvisxlV40_v40Bakedvaesd_xl_base_1.0_fp16_vaesd_xl_base_1.0_inpainting_0.1

RealVisXL-inpainting

Use Lightning models with DPM++ SDE Karras / DPM++ SDE sampler, 4-6 steps and CFG Scale 1-2

realvisxlInpainting_v5lightning.safetensors

v5Lightning

v5.0-inpainting

v4.0-inpainting

Text by @sevenof9247 <a target="_blank" rel="ugc" href="https://civitai.com/articles/3990">https://civitai.com/articles/3990</a>realvisxlV40_v40Bakedvae - inpainting<h2 id="how-to-use:-9phhbx5ao">How to use:</h2>Recommended Negative Prompt:(worst quality, low quality, illustration, 3d, 2d, painting, cartoons, sketch), open mouthor another negative promptload the modelgo to tab "img2img" -&gt; "inpaint"you have now a view options, i only describe one tab "inpaint"<ol><li>put any image there (below 1024pix or you have much Vram)</li><li>press below "auto detect size" (extention: sd-webui-aspect-ratio-helper)</li><li>draw a maske (you can resize the pencle or reset)</li><li>write a prompt, what you want to see in that mask</li><li>now you have several options</li></ol>a) Masked content<ol><li>fill and latent nothing: best suited for large changes with high denoising strength (&gt;0.8)</li><li>original and latent noise: for minor changes with Denoising (0.4 up to 0.9)</li></ol>b) Inpaint area<ol><li>Whole picture: take more the weight from the whole image</li><li>Only masked: take more weight of your prompt</li></ol>sampling steps ~40, Mask blur 4-8,After multiple rounds of testing, the suggested drawing parameter settings are:<ul><li>Steps ≥ 30</li><li>Sampler: DPM++ 2M Karras</li><li>CFG scale: 7</li><li>Size ≥ 1024x1024</li></ul><h3 id="other-stuff:-fksilbl5r">Other stuff:</h3>Merge betweenrealvisxlV40_v40Bakedvaesd_xl_base_1.0_fp16_vaesd_xl_base_1.0_inpainting_0.1

Files with SHA256: ffbdebf3ee346741490ea23192076bdcdbd53c8a8a2402d14f1ce78cef8ea534