To be clear if I generate dreams inside the 512x512 it does work, comes up with alternative inpaints. it's just on the second dream that I can't get anything outside that original 512x512 space. I've checked to make sure I'm using the right inpainting model and VAE, DDIM, I've locked and unlocked the seed, etc. Max Size 5MB or 1200x1200 Removing Object From Photo You can use AI Magic Eraser as repairing old photos, helper in cleaning faces, removing unwanted objects from a photo, and just inpaint. Adding on another overlapping "dream" like in the demo just generates black. Got it running and installed, and it generates/"dreams" the first 512x512, but even after accepting the generated image everything is limited to that 512x512 original render space. I'm struggling with what's clearly PEBKAC, but any tips would be appreciated. If I don't allow the two to touch then I can "dream" over and over on empty canvas, but if there's any overlap at all then it generates just a black image. Thanks to everyone for trying it out and my eternal appreciation to openOutpaint's collaborators :DĮDIT/UPDATE: It only generates black when overlapping with another existing render. Inpainting is an indispensable way to fix small defects. There's a moderately outdated recently updated full usage manual (and a step-by-step guide that kinda sucks) at (and respectively) No matter how good your prompt and model are, it is rare to get a perfect image in one shot. Fig 2: Image inpainting results gathered from NVIDIA’s web playground. There are a plethora use cases that have been made possible due to image inpainting. ![]() If you're using it in a colab (particularly TheLastBen's?) and are experiencing 500 errors, 404 index.html not found, etc there's a fix you can apply manually for the time being there's an issue opened on thelastben's repo for it as well but see here for details Image inpainting is the art of reconstructing damaged/missing parts of an image and can be extended to videos easily. There are several algorithms for inpainting, using different approaches for such replacement. ![]() This is achieved by replacing distorted pixels by pixels similar to the neighboring ones. You'll need to set your COMMANDLINE_ARGS value to include -api if using the extension, and set appropriate values for -cors-allow-origins if using it standalone but i can pretty heartily recommend just using the extension in most cases :) Image Inpainting is the process of reconstructing damaged or missing parts of an image. So yeah, to preface, it's vital to use an inpainting-aware model - they're specifically trained with additional unet channels and an understanding of masking.īeyond that, outpainting is handled through the dream (txt2img) tool - the img2img tool can definitely inpaint but will drop flat black on transparent pixels and is unusable for outpainting.
0 Comments
Leave a Reply. |
AuthorWrite something about yourself. No need to be fancy, just an overview. ArchivesCategories |