DALL-E Outpainting or Photoshop's Generative Fillet: Which is Better for Extending an image?

The ability to extend images to different sizes and ratios is one of the most useful features in leading AI image generators. Creatives are no longer "stuck" using the images they have. You can create new images tailored to the originals. Artificial intelligence can create a complete person from just one eye.

Adobe subscribers can access Photoshop's Generative Fill. You can sign up on Adobe Firefly or Photoshop Beta. Check out our guide on how to use Photoshop's Generative Fill Tool.

Let's start our comparison of DALL-E Outpainting and Photoshop's Generative Fill with the most extreme case; extending the frame out from a close-up image of an eye. We began with a partial image of an eye with a resolution of 1,024 x 1 024 pixels and then expanded the cropping to include eight identical-sized boxes around it.

Photoshop's Generative Fill, if set up differently (which is not possible in Outpainting at the moment), can extend the entire pupil if a single selection from the eye area is made. We used the Rectangular Marquee tool to make a selection in the frame of the eyes and then reversed the selection.

The only drawback of this method at the moment is that the maximum resolution for the entire extension around an original eye is 1,024 pixels. This creates a soft, slightly out-of-focus effect. If you zoom in you will see that the resolution is lower than the original version of 1,024 pixels.

In our second comparison, let's take a square image of 1,024 pixels and stretch it out to about twice its original size. This is more suitable for a landscape photograph. To get the best results, we played to both applications' strengths.

We extended the frame a few times to ensure that DALL-E would be able to use the pixels. The prompt was "Landscape".Outpainting is a great tool when used in this way.

Clothing and limbs are difficult to extend on people. Generative fill can be used to change clothing if you're okay with the result. Don't expect that you can just change the colors of your wardrobe. All AI image generators are currently having issues with their hands.

Let's test Generative Filling and Outpainting. Can you extend an image of a person cropped in the chest and still end up with something useful? Let's see what happens. We've used the same 1,024-pixel square and are trying to extend the image toward the legs of the man.

DALL-E's Outpainting filled in the details with a decent job. It didn't always pick up on the fact that a man had his arm raised and sometimes generated two hands in the pockets. There are four versions available and the option to generate additional hands (at the cost of credits).

Generative fill did a good job of filling in all the details. It's not perfect, but it looks passable. As opposed to DALL-E’s Outpainting you can generate new versions, then make minor corrections and not spend any more credits.

Generative Fill and Outpainting seem to extend images in a similar way, based on the examples and further testing. Each has its limitations, but it is difficult to tell which one is better. In every comparison we made, based on only a few attempts to extend images, neither one stood out as a clear winner.

Users of Generative Fill and Outpainting can benefit from the competition between Adobe and OpenAI, who are working around the clock on improving their AI systems. As the quality of image generation improves, so too will the tools and features that help us customize our images. Which app do YOU like best for extending your images?

technology news