Generative Fill for Frame Extension


Today’s Question: In your article on extending the frame of a photo using Photoshop [in the June 2023 issue of Pixology magazine] you explained how Content-Aware Fill could be used to add pixels in an area where you’ve extended the frame of a photo. Could the new Generative Fill feature be used in a similar way to extend the frame?

Tim’s Quick Answer: It is indeed technically possible to extend the frame of an image using the Generative Fill feature in the public beta version of Photoshop. However, two key shortcomings limit the effectiveness of this option.

More Detail: The Generative Fill feature available in the public beta version of Photoshop is based on artificial intelligence (AI) technology, and in some respects is rather impressive. Certainly, there is significant future potential for this technology.

While the feature is primarily aimed at filling a selected area of a photo with new details, it can also be used to extend the frame of an image. You would start by using the Canvas Size command (Image > Canvas Size) to extend the frame by the amount needed in the desired direction.

You could then create a selection of that area of extended canvas, overlapping the selection slightly with the existing image. You could then use the Generative Fill command to fill that area automatically. If you leave the text field blank when you apply Generative Fill, it will operate in a way that is similar to the Content-Aware Fill command, using the existing image to determine what should be added to the selected area.

However, there are two significant limitations to be aware of. First, the current beta of Generative Fill only fills based on an area of 1024×1024 pixels. If you select an area larger than 1024 pixels then the pixels added to fill the selection will be scaled up, which can result in significantly degraded image quality in the filled area. In addition, the results produced by Generative Fill cannot be used commercially, so you would not be able to sell or otherwise license the resulting image.

Of course, the biggest limitation of all right now is that the results of Generative Fill are quite mixed, and in some cases not very good at all. You can see some examples of that in Episode 33 of my series “Tim Talks”, which focused on Generative Fill. You can learn more about “Tim Talks” here:

Because of the limitations of Generative Fill, I recommend extending the canvas of an image by either duplicating existing pixels or using the Content-Aware Fill command, or a combination of both. This was covered in the June 2023 issue of my Pixology magazine for photographers, which you can learn more about here: