Oh, I see the problem now! You are importing StableDiffusionInpaintPipeline but then you are instantiating your pipeline using StableDiffusionPipeline, which doesn鈥檛 know how to deal with input images. You need to create your in-painting pipeline like this:
pipe = StableDiffusionInpaintPipeline.from_pretrained(model_id).to(device)
Let me know if that works 