Blurry and grainy results filling with AI

Question

Hello everyone, taking advantage of my recent indoor photography sessions, I finally had the opportunity to try my hand at AI-powered area filling. Needless to say, I am very pleased and satisfied with what has been made available. Touch-ups that used to take up to half an hour, such as removing cables or reconstructing parquet flooring, now get resolved in 30 seconds. Fantastic!

However, I couldn't help but notice some issues when I happen to select rather extensive areas of the frame (usually, I work on images that are around 25 megapixels). In these cases, the resolution of the filling suggested by Photoshop is quite lacking. The result is either grainy fillings, albeit consistent, or blurry ones. In short, they lack resolution. On the contrary, when the selection is small, the result is fantastic, perfectly detailed.

What is the reason behind this? How can I achieve sharp results even when selecting larger areas?

This is the case, for example, with panoramas where I need to reconstruct parts that literally don't exist, sometimes as large as 15% of the photograph itself.

Thank you very much 😊👋

Trevor.Dennis · Answer

At the moment, the maximum full res area that Generative Fill and Firefly can fill is 1024 x 1024 pixels. That will increase to 2000 pixels in November, but each time we use Gen Fill we will use a Generative Credit. I have not worked out exactly how this will work i.e. will the three results we get each time we hit Generate use three credits, whether we like and use the result or not? You get eith 1000 or 500 Generative Credits (GCs) a month depending on what plan you have.

https://helpx.adobe.com/nz/firefly/using/generative-credits-faq.html

You can build up a fill by selection 1024 x 1024 tiles, and even semi automate that process, but you'll be burning through those GCs.

Colin Smith shows how to automate the process on his PhotoshopCafe channel

가입하기

소셜 로그인

커뮤니티에 로그인

소셜 로그인

파일을 바이러스 검사 중입니다.

이 파일은 다운로드할 수 없습니다