Regarding your question, I think it's related to your inpainting configuration for Adetailer
From my personal experience with Forge WebUI, if you inpaint just the masked region, it will try to paint the full image into the masked region, which would create something like the image in your spoiler
If you instead inpaint the full picture, it would still only touch the masked region, but it would paint it based on the contents of that region (and the edges if you use soft inpaint)
So I would recommend that you check your Adetailer inpaint configuration