Home • Glossary • Inpainting

Inpainting

Inpainting is the targeted recomputation of a marked image area while everything outside the marking stays unchanged. It fixes single errors without regenerating a shot that otherwise works. Its counterpart, extending an image outwards, is called outpainting.

What we need it for

Generative models often deliver a shot that is right in everything but one detail: a hand with six fingers, a distorted logo, lettering that looks like type but spells nothing. For those cases inpainting is the right tool, because what works is preserved.

Where the limits are

On a still, inpainting is straightforward. In motion it gets demanding, because the fix has to stay stable across every frame or the area flickers. Where the repaired surface turns or becomes occluded, a fresh generation is often faster than the repair.

So we decide case by case: repair or regenerate. How often we regenerate is described under AI film, where the ratio of waste to usable output is set out.

Frequently asked questions

What is inpainting?

Inpainting means replacing a marked image area with newly computed content while the rest of the image is preserved. In video production it serves to correct single errors, for instance on hands, type or product detail.

What is the difference between inpainting and retouching?

Both replace image content. In classic retouching a person works with tools such as clone and brush, in inpainting a model computes the content from the surroundings and a description. In practice we combine the two.

Can you see inpainting in the finished film?

Not when done carefully. The giveaways are soft edges at the repair and flicker across frames. We check both at full resolution and frame by frame before delivery.

Related terms

Read on: All terms in the glossary • Image-to-video

Does your product have to be exact?

Labels, logos and shapes get corrected by hand. Show us the product in the first call.

Book a first call