Automatic image
masking and blending.

Object detection turns “the people in the background” into located subjects. Segmentation follows their silhouettes down to the pixel, giving Rascal a precise mask without the hand tracing.

Rascal gives the edit room to breathe: expanding the mask for shadows, closing small gaps, and carving out protected areas. Gaussian feathering softens the boundary so the replacement can settle into the scene.

Once your fill model paints the replacement, Rascal blends it into the shot pixel by pixel. The mask controls how much of each image appears along the boundary, while protected details stay exactly as they were.

Every pixel gets a clear job: open to change, part of the blend, or off limits. Rascal turns your product selection into a protected region, keeping the details that make it yours.

Inverting the selection makes the surroundings editable. Feathering can soften the transition, but protected pixels always take priority—even where a blend would otherwise reach them.

A new setting, with your actual product still in the shot. Rascal blends the generated background through the mask and copies protected pixels unchanged. The lettering survives because it is never regenerated.

A rough stroke is enough to get started. Rascal reads changes in brightness across it to find the tear’s edges, then pulls the mask in around the damage. Where the edges are unclear, it keeps your original stroke width.

Rascal measures each pixel’s distance from the repair and calculates a Gaussian fade. That gives the new pixels a soft landing in the photograph, without a hard cut around the fix.

For fills with a tonal mismatch, Poisson compositing preserves the fill’s local gradients while matching the original image at the boundary. Weighted blending then fades the repair into place.