CivArchive
    Utopian Inpainting workflow - v2.2
    Preview 120255368

    Over view

    This is a simple workflow designed to closely reproduce the inpainting features available in WebUI. And specifically designed for use with my model, Utopian Inpainting.

    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, Epic Realism 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"

    -Basic feauture-

    • Inpainting with refiner model

    • Inpainting full image and crop image select

    • Pre-resize image with scale by % and long_side resize select

    • Basic K-sampler and Lanpaint support

    • Manual mask and sam2 Auto masking select

    • Optional Vae and Clip selector

    • Postprocess Mask detailer (testing)

    Upcoming Update plan: Outpainting Support, More Accurate Auto masking.

    -How to use-

    Resources you need:

    Settings:

    [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.

    • crop target size
      Scales the size of the crop from the image.

    Cropped_image = Original_img * Pre-resize * crop_target_size
    • crop mask extend
      How much of the image to crop from the mask.

    [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)

    • Detailer denoise
      recommended 0.3 ~ 0.5

    • 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.

    Description

    1. 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."

    1. Fixed an issue where vae was not switching correctly when loaded.
      Inpaint model and refiner model must have the same vae compatibility

    FAQ

    Comments (7)

    warriors666Feb 7, 2026
    CivitAI

    where is the denoise?

    habibing
    Author
    Feb 7, 2026

    Not in the seed column?

    warriors666Feb 11, 2026

    @habibing How do you adjust your outpainting in your workflow? For example, starting from a 512x512, I want to create a 4:3 aspect ratio. Please?

    habibing
    Author
    Feb 12, 2026

    It's nearly impossible to output an image exactly to a specific aspect ratio. The sdxl VAE requires a factor of 16 for both height and width, and it's too complicated to calculate how long to stretch the image for outpainting to meet a specific aspect ratio and still be a factor of 16.
    First, your 512x512 image will be rescaled by specifying the long edge or the scale factor. SDXL does not support sizes below 1024, so we'll assume you specified 1024 as the long edge. If you set the left outpaint slider to 0.2, the left side of the image will be extended by 1024*0.2, meaning the image width will be 1024*1.2 (1228), resulting in a width of 1228 x height of 1024.
    If you want to resize an image to 4:3, you can calculate the width needs to be 1365 based on the height of 1024 and the 4:3 ratio. Dividing 1365 by the width of 1024 gives us 1.3330078125, so we can see that the left or right slider needs to add up to around 0.33.

    chero90Feb 8, 2026
    CivitAI

    tryed to install it on a stock- comfyUI (downloadend json and dropped it in). I'm getting some error about unknown nodes.

    In addition my node manager gives an info line (but w/o Context): "Some extensions are disabled due to incompatibility with your current setup [....]. Its the Impact Subpack and Brushnet with this issues.

    habibing
    Author
    Feb 8, 2026

    Does this mean that some nodes can't be installed from the manager? In that case, you can get the zip file directly from GitHub and extract it directly to ComfyUI\custom_nodes.
    Go to the GitHub page, click the green "CODE" button, and download it from Download ZIP.
    Since it's a small node, I don't think there will be any dependency issues.
    ComfyUI-Impact-Subpack
    ComfyUI-BrushNet

    Or you may be able to find it by setting the channel to dev in the manager settings.

    habibing
    Author
    Feb 8, 2026

    For example, if you manually install impact subpack, the file structure will be as follows:
    ComfyUI/custom_nodes/ComfyUI-Impact-Subpack-main

    Workflows
    Illustrious

    Details

    Downloads
    472
    Platform
    CivitAI
    Platform Status
    Available
    Created
    2/7/2026
    Updated
    5/14/2026
    Deleted
    -

    Files

    utopianInpainting_v22.zip

    Mirrors

    HuggingFace (1 mirrors)
    CivitAI (1 mirrors)