피파 한 줄 정리: Outpainting은 frame 밖으로 확장. 작은 step으로 여러 번이 한 방에 크게보다 훨씬 안전해. Aspect ratio 변환·panoramic expansion에 강함.
Mental model: You have a beautiful 4x6 photo, but you need a 16:9 banner for your website header. Instead of cropping (losing content) or stretching (distorting), you could ask an artist to continue painting the scene beyond the original edges — extending the sky, continuing the landscape, adding more foreground. That's outpainting: generating new content that seamlessly extends an existing image beyond its original boundaries.
How Outpainting Works
Outpainting is conceptually similar to inpainting, but instead of filling a hole inside an image, you're filling new space around it. The model sees the existing image as context and generates new content along the extended edges that matches the style, perspective, lighting, and content of the original.
Original Image Outpainted (extended right) ┌──────────┐ ┌──────────────────────┐ │ Mountain │ │ Mountain + more sky │ │ Scene │ → │ Scene + trees │ │ │ │ + meadow │ └──────────┘ └──────────────────────┘ 4:3 ratio 16:9 ratio (seamless)
Common Use Cases
- Aspect ratio conversion: Convert a portrait (9:16) to landscape (16:9) for different platforms — Instagram Story to YouTube thumbnail without cropping the subject.
- Adding headroom: A portrait that's too tightly cropped? Extend upward to add space above the head.
- Panoramic expansion: Extend a landscape left and right for a wide banner or wallpaper.
- Scene completion: Your generated image cut off something interesting at the edge? Extend to reveal more.
- Background extension: Need more background for text overlay in a marketing layout? Extend the less-detailed areas.
Challenges and Limitations
Outpainting faces some unique challenges that inpainting doesn't:
- Diminishing context: The further you extend from the original image, the less context the model has. Large extensions can start to look generic or disconnected.
- Perspective consistency: The model must maintain correct vanishing points and perspective lines as it extends. Slight mismatches create a subtle "uncanny" feeling.
- Lighting continuity: Shadows, highlights, and color temperature must flow naturally across the boundary. Mismatched lighting is one of the most common outpainting artifacts.
- Content invention: Unlike inpainting (where the model fills a gap with contextual clues from all sides), outpainting only has context from one direction. The model must invent plausible content with less information.
Extending a tight portrait by 200% on all sides in a single pass → Background becomes generic, perspective breaks, lighting doesn't match
Extend 30% to the right, then 30% to the left, then 20% up, with a prompt matching the scene's style at each step → Seamless, context-aware expansion
- Outpainting extends an image beyond its borders by generating new contextually matched content.
- Most useful for aspect ratio conversion, adding headroom, and panoramic expansion.
- Extend in small increments for best results — diminishing context degrades quality.
- Works best with gradual, textural areas; struggles with complex structural content.