Just a random little "tip" I want to throw anyone learning this.
retained text
I'm having to edit a whole lot of pictures right now, small little changes over and over. If you're doing this then there is no reason to run the entire image through a model. There are node packs like AusBoss which has a crop for inpaint and stitch inpaint node.\\ Basic WF would be Load Image > Image and mask > Crop for Inpaint > whatever you're loading the image into, something like Qwen > Image comes out of that into Stitch Inpaint > Image comes out of Stitch Inpaint into save image. You also need to connect the "stitcher" from Crop For Inpaint to "stitcher" in Stitch Inpaint. You will also need to add in vae encode/decode depending on how you're using the image, if it is being used as a latent. This can greatly reduce the size of the image you're generating depending on the size of the mask. Oh, almost forgot that. You do need to create a mask in Comfy so the Crop For Inpaint node knows what part of the image to crop out. You're prompts will also need to keep this in mind. You're no longer prompting for the entire image, you're prompting for the area that would be cropped out. I tagged this as a tutorial but this is a pretty bad tutorial. If what I said doesn't make sense, just ask or look at the example workflows in the ComfyUI-AusBoss node pack.
document fetched 2026-10-15 00:45 · canonical https://reddit.com/r/comfyui/comments/1x5t61c/just_a_random_little_tip_i_want_to_throw_anyone