Facebook has released https://github.com/facebookresearch/segment-anything, which would be a boon for advanced workflows such as detecting and extracting small, complex details from an image, upscaling and re-diffusing them, and then blending them back into the original image.
The example code seems simple enough to adapt, and there are multiple model sizes to choose from (91M, 308M, 636M).
Facebook has released https://github.com/facebookresearch/segment-anything, which would be a boon for advanced workflows such as detecting and extracting small, complex details from an image, upscaling and re-diffusing them, and then blending them back into the original image.
The example code seems simple enough to adapt, and there are multiple model sizes to choose from (91M, 308M, 636M).