Magic Fill
Intelligently fill in missing or damaged areas of your images using advanced AI inpainting technology.
Overview
Magic Fill is an AI-powered inpainting tool that seamlessly repairs, removes, or replaces selected areas of an image. By analyzing surrounding pixels — color, texture, lighting, and composition — the model generates new content that blends naturally with the rest of the photo. Whether you are restoring an old photograph, removing an unwanted object, or extending a background, Magic Fill delivers professional results in seconds.
Key Benefits
- Seamless Restoration: Repair damaged or missing areas with natural-looking results that match the surrounding image.
- Object Removal: Erase unwanted people, objects, or text while the background fills in automatically.
- Background Extension: Expand canvas edges to fit different aspect ratios without distorting the subject.
- Creative Enhancement: Replace or modify image regions to explore new compositions and ideas.
Quick Start
Follow these steps to make your first Magic Fill edit:
- Upload your image or open an existing asset from your library.
- Choose a selection tool — brush, lasso, or rectangular marquee — and paint over the area to fill.
- Optionally type a short text prompt describing what should appear in the selected region (e.g., "clear blue sky", "wooden floor").
- Click Generate. The AI produces one or more fill candidates within seconds.
- Preview each candidate, accept the best result, and export or save to your library.
Pro Tip
Extend your selection a few pixels beyond the unwanted area to give the model enough surrounding context. A tight selection that barely covers the target often produces visible seams at the edges.
Selection Tools
Magic Fill provides three complementary selection tools so you can precisely define the fill region regardless of shape or complexity.
Brush Selection
Paint freehand over the target area with an adjustable brush size. Best for organic shapes, hair, foliage, and irregular objects that don't have clean straight edges.
Lasso Selection
Trace the outline of the target area with click-to-place anchor points. Ideal for removing objects with distinct, predictable borders such as cars, signs, or furniture.
Rectangular Marquee
Draw a rectangular selection for fast, uniform fills. Use this when extending a background, replacing a sky, or filling a large uniform region.
Smart Selection
Click a point in the image and let the AI detect the object boundary automatically. Speeds up removal of clearly defined subjects like people or products.
Edge Expansion
Outpaint beyond the image border by selecting the canvas edges. Perfect for converting portrait crops to landscape without losing the original subject.
Non-Destructive Masking
All selections are stored as masks — the original pixels are never permanently deleted until you explicitly apply and export, keeping your workflow reversible.
Common Use Cases
| Use Case | Recommended Tool | Typical Result Time | Tips |
|---|---|---|---|
| Remove person from background | Smart Selection or Lasso | 5–15 s | Include a 10 px border around the subject |
| Fix scratches / damage on old photos | Brush | 3–10 s per patch | Work in small patches for better accuracy |
| Extend background / outpaint | Rectangular Marquee | 10–30 s | Add a prompt for best coherence ("rolling hills", "city skyline") |
| Remove watermark or text overlay | Brush or Rectangular | 5–15 s | Select slightly larger than the text bounding box |
| Replace product background | Smart Selection + outpaint | 15–40 s | Combine background removal with a fill prompt |
Best Practices
Selection Quality
- Keep your selection 5–15 pixels larger than the area you want replaced so the model can read edge context.
- Avoid extremely large selections (>50% of image area) in a single pass — split them into overlapping regions for smoother blending.
- For fine details like hair, use the Brush tool at 80–90% opacity rather than hard edges.
Prompt Writing
- Keep prompts short and descriptive: "grassy field, afternoon light" works better than a long paragraph.
- Mention lighting conditions if they differ from the rest of the image (e.g., "overcast sky" instead of "blue sky").
- Leave the prompt blank to let the model infer content purely from surrounding pixels — this works very well for background repairs.
Iterative Editing
- Generate 3–4 candidates and compare before accepting. Even small variations can meaningfully improve the final result.
- If the first pass has a faint seam, run a second narrow-brush pass over the seam area only.
- Use the non-destructive mask until you are fully satisfied, then apply and export at the highest resolution needed.
Common Mistakes
- Selecting too tightly around the target — leave a pixel buffer so the model sees the boundary context.
- Using prompts that contradict the surrounding scene (e.g., "bright sunny sky" in an indoor photo).
- Attempting to fill very high-frequency detail areas (chainmail, fine text) in a single large selection.