CivArchive
    ← All articles
    Published October 15, 2025by Snakehands

    My workflows

    194 views1 reactions2 comments on CivitAI4 collected
    workflows

    Some people have asked about what workflows I use in ComfyUI. Here are the main three (I also have one for upscaling, but it's similar). The main custom nodes I use are rgthree, efficiency nodes, and LayerUtility. There's nothing crazy here, but hopefully this can be helpful to people just starting out with ComfyUI.

    T2I + I2I + CtrlNet

    Basic generation with the control net for sketch-to-image generations. Toggle the latent switch to switch between text2img and img2img. Bypass the control net if you don't need it.

    When using control net, first I get an inverted (white on black) png of my outlines with the same aspect ratio and at least the same resolution as the image I want to generate. These can be very messy, but I get better results by using bold outlines. Using Xinsir Scribble or Canny (available on huggingface), I start around 0.7 strength/0.7 end_percent with slightly lower CFG than I would usually use for the model. Then I adjust those three values until I get something I like. I usually end up taking 4-5 images into an image editor and splicing them together to get all the key features I want, since the generator will rarely give a perfect image. Then I just pass the Frankenstein image through a light Img2Img to clean it up before inpainting.

    Inpainting

    Basically what this does is enlarges the part you mask on the image to 1024x1024 so you can do a "full-power" generation on just that part without losing any detail. Then it shrinks it again and stitches it back into the image. At denoise >0.3 or so it usually leaves seams that have to be dealt with somehow (I smooth them manually, but I've been meaning to add some kind of feather effect to the edges) Fixed seams

    Inpaint+CtrlNet

    Small features in my sketch rarely come out in the initial generations, so if there's a small feature that I really want to include I fire this one up. It does the same thing as the Inpainting workflow, except it does it to the sketch (from the initial T2I control net) in parallel and uses the cropped sketch as a control net for inpainting.

    Attachments