<h1 id="over-view">Over view</h1>This is a simple workflow designed to closely reproduce the inpainting features available in WebUI. And specifically designed for use with my model, <a target="_blank" rel="ugc" href="https://civarchive.com/models/838736?modelVersionId=2302750">Utopian Inpainting</a>.Update: Changed sam2 to sam3 Old update: Added Openpose, Detail daemon, fill mask area, basic Adetailer, and more. I realized the pure SDXL models is good as a refiner. After testing several options,<a target="_blank" rel="ugc" href="https://civarchive.com/models/277058/epicrealism-xl?modelVersionId=2514955"> Epic Realism</a> delivered good performance with both high-resolution and low-resolution images. Set the sampler 2 configuration to "dpmpp_2m_sde" "karras" and try adjusting the "refiner ratio" to around "0.6 to 0.8"<h2 id="basic-feauture">-Basic feauture-</h2><ul><li>Inpainting with refiner model</li><li>Inpainting full image and crop image select</li><li>Pre-resize image with scale by % and long_side resize select</li><li>Basic K-sampler and Lanpaint support</li><li>Manual mask and sam2 Auto masking select</li><li>Optional Vae and Clip selector</li><li>Postprocess Mask detailer (testing)</li></ul>Upcoming Update plan: Outpainting Support, More Accurate Auto masking.<h2 id="how-to-use">-How to use-</h2><h3 id="resources-you-need:">Resources you need:</h3><ul><li>Inpainting Model: <a target="_blank" rel="ugc" href="https://civarchive.com/models/838736?modelVersionId=2302750">Utopian pony inpainting </a>or Any SDXL Inpainting model</li><li>Refiner Model: Your favorite models</li><li>Optional Clip: <a target="_blank" rel="ugc" href="https://civarchive.com/models/1784001?modelVersionId=2021051">Illustrious_base_clip_L</a> and <a target="_blank" rel="ugc" href="https://civarchive.com/models/1784001?modelVersionId=2023635">Illustrious_Base_CLIP_G</a></li></ul><h3 id="settings:">Settings:</h3>[Optional loader] Select if you want to use Clip or VAE instead.[Inpaint mode] -Normal Normal Inpainting. -Refiner Inpainting with refiner. -Lanpaint Use Lanpaint, The output results are different from Ksampler. It is difficult to say which is better.[Resize method] -percentage Pre-resize image with scaled %. -Long side Pre-resize the image based on the long edge.[Inpaint area] full Inpaint with pre-resized image size. -crop Crop pre-resized image according to mask.<ul><li>crop target size Scales the size of the crop from the image.</li></ul><pre><code>Cropped_image = Original_img * Pre-resize * crop_target_size</code></pre><ul><li>crop mask extend How much of the image to crop from the mask.</li></ul>[Mask method] -Manual Right-click on the load image and select "Open in mask editor" to manually draw a mask. -Auto A mask is generated automatically; use the mask prompt to define the area you want to mask.[Postprocess] -After Detailer Enhance the Inpainted image by running a mild inpainting pass after finished. -with refiner Also use refiner model when Detailer. (With refiner doesn’t perform well at the moment)<ul><li>Detailer denoise recommended 0.3 ~ 0.5</li><li>Detailer zoom Set how much of the surrounding area is included when cropping. The higher the value, the more area around the mask will be cropped.</li></ul>

Utopian Inpainting workflow

<ol><li>Changed Sam2 to Sam3 I have confirmed that there is a compatibility issue with Transformers 5. The model is downloaded to ComfyUI/models/sam3. It is 3.2GB and may cause OOM if VRAM is low. This is a simplified version, so it only responds to single words such as "hair." It does not support expressions such as "hair, boots" or "hair and boots."</li></ol><ol><li>Fixed an issue where vae was not switching correctly when loaded. Inpaint model and refiner model must have the same vae compatibility</li></ol>

Files with SHA256: 1626287c58f5a198eb03320db2c34a7bb1e918695914cf5f08fcb15e1d84cc25