I made that comment originally 9 years ago, and since then I have started to simplify it a bit. It's essentially the same technique, but I do it the following way:
Select a smooth area.
High Pass (stamp visible or flatten first if needed).
Define Pattern.
Undo the last steps.
Use the pattern with Linear Light blend mode at 50% fill. Use Fill% to control strength (not opacity unless it's a layer style and you have no choice).
The main change is that I don't copy it over into a new document for editing before defining the pattern. Saves time. For grain I find the editing isn't needed a lot of the time (but if you experience seams, use Offset to locate and fix).
We could have kept it private, but then users didn't understand why (wouldn't see that post), and old content that users find useful would not be reachable (this is a tradeoff though).
1024x1024 is sufficient for smaller areas, and it might work better to do it in sections if the resolution is an issue? But yeah, resolution is a major problem for professional work.
Anyway, i think the generative fill beta is a great proof of concept. Very user friendly compared to the SD plugins out there. And also very useful in situations where content-aware is insufficient.
Quality-wise, it is not the best, just ok. This is what they need to focus on going forward. Midjourney is leagues ahead there. Feature-wise, i guess SD wins - but it can be complicated as hell to use. 😅
Photoshop
Hot