Difforum · Fill Reveal (AI)

Complete what the camera reveals, with an image model (AI hole fill). Guide Frames, H3 Shot and Keyframes mark the area the camera uncovers (outside the original picture) in a mask. This node repaints only that area by inpainting, so edges become new, coherent scenery instead of gray or stretched pixels. The known pixels are kept exactly. Works with any image model through ComfyUI's core `InpaintModelConditioning`: a dedicated inpaint model (SDXL inpainting, Flux Fill) gives the cleanest seams, a regular checkpoint works too. Run it only on the frames a video model will see (first / last frame, keyframes) - it is one diffusion per frame.