outpainting

Outpainting is an image synthesis technique that extends a picture beyond its original borders by generating new pixels in the surrounding area. Unlike inpainting, which fills gaps inside a frame, outpainting aims to preserve and extend existing geometry, lighting, and texture as the scene grows.

Modern outpainting relies on generative models such as diffusion systems, autoregressive networks, or generative adversarial networks. The user provides the image and defines an extrapolation region, often with a mask or by choosing a direction. The model then integrates the border cues to extend color, perspective, and detail.

Techniques include edge-aware extrapolation, multi-step generation, and blending to improve coherence with the existing frame. Some workflows use an initial pass to extend the scene, followed by refinement passes to align texture and lighting across boundaries.

Applications span creating wider panoramas, expanding concept art for films and video games, and assisting visual storytelling or restoration when extending a frame is desirable. It may also be used to explore variants by imagining what lies beyond the current view.

Limitations include possible geometric or lighting inconsistencies, and the risk of hallucination where generated content does not match real-world context. Ethical considerations involve authenticity, copyright, and disclosure of AI involvement to prevent misinformation.

Evaluation is typically qualitative, focusing on realism and cohesion, though researchers may use perceptual metrics and user studies to assess continuity across borders.