The team cannot wait to see what you've created with this thrilling, new, beta technology!
This is the FUN sharing post.
We've been having such a blast exprimenting and playing for the past few months, it's so exciting to have you join in the joy!
Show off those creations! Bonus if you also upload the orignal image and the prompts.
Ready! Set! GO!
Thank you for trying out Photoshop (beta).
P.S. Provide Feedback and help shape Generative Fill: If you would like to provide feedback on the overall experience, report any bugs, or suggest new features please let us know in the beta community announcement.
If you prefer Discord, you can find us there too!
90
Replies
My prompt was simply "Black leather Jacket."
Here's a first pass at doing an infinite pull out.
Sometimes I think I'm being really creative.... this might be one of those times. No, I'm not sure about the text- but I'm hungry and thinking "Man a couple of eggs would go down good right now...."
Often times I will enter "black leather jacket" and it produces a different black leather jacket each time.
This time I entered stuff like "black rockabilly jacket" or "black sweater" to get what you see here.
My son looked so cute and pensive in this picture from 2005, but the background was busy and annoying. I cycled through a few options and got this result using "sand and lake with out of focus trees in the background at sunset."
Of course, now I'm noticing a weird thing behind his right elbow, but hey.
as a professional 360° photographer i often need to retouch panoramic images after the stitching is finished. usually, this is limited to removing the tripod and its shadow in full spheric panoramas, and sometimes some moving people.
recently i did a (beginner's) mistake i was very sorry about: during a panorama shooting i needed to swap my memory card and after this i resumed my capture at the wrong image in my series (instead of better starting over).
later in the office, the disappointment was huge as this was a place i won't come back to very soon: 2 images out of 16 were missing 😞
as the images for the panorama were taken with a wide-angle lens the hole in the panorama was pretty large and it went to the collection with failed stuff.
when i heard about the generative fill, i tested it and then remembered this panorama. after fiddling around with the best way to fill this huge missing area in the panorama i ended up in using a 1024x1024 pixel square selection to iteratively fill the hole with 13 generative patches.
hopefully, this feature will be able to handle larger resolutions in the future!
so the new generative fill literally saved this beautiful panorama from the trash can.
after this very pleasant experience i tested the generative fill with a routine task: eliminating the tripod and its shadow. i choose just the same panorama for this.
after 10 generative fill layers, i was simply amazed about the visual quality this method provides compared to the healing brush or using the clone stamp tool. no annoying repetitive cloning patterns anymore, vivid, believable structures and lighting!
besides all creative applications, the generative fill feature is a huge time saver and enhances the quality of professional 360° retouching significantly.
feel free to see this panorama interactively in full 16000 x 8000 pixel resolution here: https://panobu.ch/aipatchedpano
.png){kind=avatar}
Learning more about using generative fill and how it's behaving
after the panorama i've saved from trash can with filling the hole of 2 missing images with the new generative fill (i posted it here a few days before), i tried another one.
this time it's a sloppy made handheld cylindrical pano taken with a sony a7R (36 MP) and a 50 mm lens. the unstable weather and a lot of miles left to hike made me hurry and so it's a pretty incomplete panorama with large missing parts in the cloudscape and in the foreground (we'll do this later in post ... 😄 ).
as the total size of the pano is about 66k x 11k pixels the resolution limit of 1024 x 1024 pixel for the generative patches made it a bit tedious to fill in all missing parts one after another. but this step-by-step-process also allows for some control over the process as you have always 3 suggestions presented to you by the generative fill engine. another control you have is the edge where you begin to add new generated content.
some points i found out during this test:
- it's always better to have 2 (or even better 3) sides of existing content for a new 1024 x 1024 square.
- if you've larger holes of missing edge areas start from several points/sides
- make always sure your selection includes 10 to 50 pixel wide egdes of existing content.
- also, make sure you're adding up content towards transparent areas. a black or white background isn't working.
- the offered generative content doesn't have always the same image quality. sometimes i noticed less sharp suggestions. so it's recommended choose from best match and best quality.
- for (very) large images it may be an advantage to work with a higher resolution and scale down the result.
- in cloudscapes the generative content obviously tends to shift the hues of the blue sky in an unwanted direction. some color correction on such parts may apply.
- in cloudscapes you can exceed the fill area limit if the structure is soft by nature.
- the same applies in soft, out-of-focus areas in the foreground
again i was absolutely amazed about the AI generated content suggestions made based my the existing photographic content.
the result (scaled down from 66k to 48k width) can be viewed here: https://panobu.ch/drusentor
fun fact: the panorama is taken exactly on the border between austria and switzerland. can you spot the #1 border marking made in 1937? 🙂
AdChoices
