later that turned into creating things from nothing because it was so good at denoising that if you give it chaos it can make a fine art style painting or a photograph of a tree.
#DESCARGAR INPAINT FULL HOW TO#
sorry, you can probably experiment more now that I gave you a crash course.Ĥ) as far as I understand at the core denoising is basically all SD does (SD comes packed with 30+ other ML things and tools that make it useful I guess/think), it was trained by being fed known images with a known about of noise added so that it learned how to fix messed up images. I use this for turning my friends into zombies using a low denoise and high CFGĬ) latent noise replaces the masked section with true noise that then gets used to generate a new image, I would likely use this for your hat example if there wasnt a mspaint hat added, at the most basic SD can only remove noise from an image, all other settings are to control how that noise is used.ĭ) latent nothing I think I only clicked on once, I dont remember what it did but I dont think it was helpful. this would be good if you wanted to remove a person from a room then use SD to recreate the wall in the background.ī) original, feeds the cropped portion directly to SD for use with your prompt and other settings. This can get you extra details, but again, RNG wins in the end.ģ)A) fill - before running your prompt it removes the masked image portion and tries to "fill" it with the colors that touch the edge of the crop to poorly "hide" the removed thing and then runs the prompt.
#DESCARGAR INPAINT FULL FULL#
when you inpaint at full resolution all that means is in your hat example if you change nothing else it basically crops the image to the dogs head, makes a cool or HORROR of a hat, then scales it down and drops it back in the original. Inpaint padding is similar but for when you inpant a small part but use the inpaint at full resolution option. case by case Trial and error lives here it seems.
You can get lucky with prompts alone but I've used raw generation and assisted both a lot now and assisted is much better and faster if I want something specific and not just to dance with the tech.Ģ) Mask blur sometimes helps merge the inpainting and the unchanged part around it. The pro tip move here is use photoshop/mspaint to make a crap drawing of a hat on the head then use img2img to make it "real". any time you "talk" to SD it wants to make you one image, so you need to ask carefully. I'm not good at it but I try to list my steps.)ġ) with the hat shaped example you would just type about the hat. (Check my post history for examples of img2img I've tried to help other folks with. Disclaimer, I'm just a guy on the internet, I learned by trial and error, I have a fast PC and thats all my qualifications.
0 kommentar(er)
