Inpaint Outpaint
These features available from version 0.2.4
Inpainting and Outpainting is available in the img2img tab. The image is preferred to be a generated image, but works with external images too if the image is resized to compatible resolutions.
SD1.x: https://huggingface.co/webui/stable-diffusion-inpainting/blob/main/sd-v1-5-inpainting.safetensors
Of course, you can use another models too, but the images may will be inconsistent.
- white - the allowed area to modify the original image
- black - the forbidden area, SD will not modify the black parts of the mask on the original image
- grey - the grey areas is the part of the initial image, where there is no "drawed" image content on the initial image. The mask image need to overlap these regions with white
Image inpainting is means to mark one or more place on the image. This image parts will be modified under the diffusion. Eg. an allowed area where the AI allowed to work knowing the other parts of the image.
In the GUI, after loading the base image, you can select three brush shape (circle, rectangle or triangle):
On the queue list in the Jobs & Images tab, you can select a generated image, right click on it, and select "Send image to img2img". This will load the image into the paint area on the img2img tab.
Alternate, you can use the file selector on the bottom of img2img tab:
This will open any image file without checking the source image resoulutions.
If the loaded image is larger than a normal SD image resolutions, you can resize it in the paintig area using the image resize button on the top:
Original sd.cpp PR #511