Yes. Inpainting workflows accept any JPG or PNG regardless of how it was generated. Upload a Midjourney output, a DALL-E image, a Stable Diffusion generation, a Flux image, or a photograph. The inpainting model works on all of them.

Most inpainting workflows include a built-in masking tool. Right-click on the Load Image node to access the masking option. Paint over the area you want to fix. The mask should cover the problem area plus a 10-20 pixel margin for blending.

It depends on how much you want to change. Subtle fixes like skin texture or minor blemishes: 0.3-0.5. Moderate changes like object swaps or face correction: 0.5-0.7. Major replacement, completely different content: 0.7-0.9. Start at 0.5-0.6 for most tasks.

The most likely cause is denoise set too low. At values under 0.3, the model does not generate enough detail to match the sharpness of surrounding pixels. Increase denoise to at least 0.5. Also make sure you are inpainting at full resolution, not a downscaled version of the image.

Yes. Distorted hands are the most common AI generation defect and a primary inpainting use case. Mask the entire hand (not just the fingers), write a specific prompt ("human right hand, five fingers, relaxed natural pose, matching skin tone and lighting"), and set denoise to 0.6-0.7. Expect 2-3 attempts. Hands are hard for AI.

You can mask multiple areas in a single pass, but results are usually better when you fix one area at a time. Each inpainting pass is cleaner when it has a single objective. Fix the hand first, save the result, then fix the texture, then remove the unwanted object.

Soft inpainting uses a gradient mask instead of a binary on/off mask. The center of the mask gets full replacement. The edges get gradual blending with the original content. This produces the most natural-looking results because there is no hard boundary between original and regenerated pixels. In ComfyUI, soft inpainting is achieved through mask feathering nodes or by creating gradient masks manually.