Photoshop Generative Fill - YAWN
Vložit
- čas přidán 23. 05. 2023
- In this video, I demo how to use the new feature Generative Fill found in the latest version of Photoshop Beta.
Please subscribe to my newsletter!
anthonymorganti.substack.com/...
Check out one of my newer websites - The Best in Photography:
bestinphotography.com/
Please help support my CZcams channel - consider purchasing my Lightroom Presets:
www.anthonymorganti.com/
To get more info about Photoshop, go here:
prf.hn/l/lGnjDBl
Here is the list of my recommended software, along with any discount codes I might have:
wp.me/P9QUvD-ozx
Here is a list of my current cameras, lenses, etc.:
wp.me/P9QUvD-ozG
Help me help others learn photography. You can quickly offer your support here, where I receive 100% of your kind gift:
ko-fi.com/anthonymorganti
You can change the default amount to the amount you want to donate.
In this video, I demo how to use the new feature Generative Fill found in the latest version of Photoshop Beta.
Please subscribe to my newsletter!
anthonymorganti.substack.com/subscribe
Check out one of my newer websites - The Best in Photography:
bestinphotography.com/
Please help support my CZcams channel - consider purchasing my Lightroom Presets:
www.anthonymorganti.com/
To get more info about Photoshop, go here:
prf.hn/l/lGnjDBl
Here is the list of my recommended software, along with any discount codes I might have:
wp.me/P9QUvD-ozx
Here is a list of my current cameras, lenses, etc.:
wp.me/P9QUvD-ozG
Help me help others learn photography. You can quickly offer your support here, where I receive 100% of your kind gift:
ko-fi.com/anthonymorganti
You can change the default amount to the amount you want to donate.
The EXIF data wouldn't support a photo of the kids at/in the bar, relax.
In your 1st example you need to expand the selection by a few pixels or it's only got a person as the reference. If you had followed the instructions you would have had fewer issues.
I was going to say that
That was my first thought as well - expand the pixels and see if it worked better.
Out of all the people on CZcams, you are the only one who has trashed it. I have been using it and it has done a decent job. You don't select the entire sky to add clouds. Maybe you should take a closer look at the people who use Photoshop and have put videos up on how to use it. If you want to have a cloudy sky, why not use sky replacement? It seems you have intentionally gone out of your way to knock the product. It's a beta version, meaning the AI is learning itself and given time it will be a total game changer. If you want to inverse a selection, there is a button you can press on the contextual task bar. I suggest you try to use it a little more, because you have made videos of other software and made mistakes, because you did not know how to fully use the software. I look forward to a follow-up video, hopefully you have fully grasped the fundamentals of the software then.
If you follow the instructions and make the right selections it’s working very well……
I usually agree with everything you say, but this seems kinda contrarian. 90% of the time I've used it, it's worked amazing. It really is game-changing, makes Ai more accessible, and is an incredible evolution of the Content Aware Fill.
Concentrate on the positive. What it can do, which is amazing. Why would you ever want to remove the main subject of an image?
Thanks Anthony, I’m a painter as well as a photographer and have been planning an image that represents the Australian eucalypt forest landscape. I had a layer with tree icons in the foreground and tried many different ways in photoshop to manipulate a photo of the bush to the image I had in my mind for the background. I ended up conceding defeat. A few days later I saw an ad for the beta version of photoshop, downloaded it and opened a new blank layer, described the image in my head hit the button and was absolutely thrilled to see exactly what I had imagined. I needed only to desaturate it. Printed it on canvas with deletions fot the tree icons which will be painted in and the background painted over with layers of washes. I willl send you a copy of the finished piece in ,probably,August.
My understanding from watching others use it, they select more outside the selection. I think the AI tries to blend the selection with the background.
Thanks for another video. I have found when you use a loose selection with the lasso you get pretty good and consistent results. When you use the select subject tool I think the algorithm is trying to replace the subject and you get weird results.
Thanks for the heads up, Anthony!
I found it worked famously in adding more canvas to an image . Crop to wanted size then mark out the extended size and fill ...boom done !
Awesome Tony! Thanks for the review 😀
Thank you, Anthony. I'm sure influencers will be "traveling" the world from behind their desk!
Thanks for your honest opinion and demo of this new feature. Right now I'm thinking I'll see.
I've used it a little. I think the technology so far is both amazing and scary. I do a lot of compositing and I can see using it if I can't find an element I need at the perspective I need it. That said, if I use it instead of an image I take, then the composite is no longer all mine. Sometimes, like the sky, it might be easier to just replace the sky rather than generate several times. It will get better, and it is here to stay, so I'm going to keep at it.
thanks, some videos I've watched where clouds work are where you have a smaller selection in the sky (i.e. 2/3 of selection far from border).
It appears that we aren't too far away from just generating photos instead of shooting them.
Hi Anthony. I general I'm not a fan of the availability of generative fill AI due to some of the reasons you've mentioned. Having said that, I think it could have some benefits when trying to fix images. For instance, I recently was able to get some shots of a Great Blue Heron flying off with a good size trout in his beak. Unfortunately, a couple of the better images from a sharpness and composition standpoint had issues where I was leading the heron a bit too much in the panning and ended up cutting off part of its feet. So, I'm curious to see if in a case like that I could extend that side of the image and have it fill in what should ideally be the rest of its feet and the background. I haven't had a chance to try it yet. On a PS Beta related front, I've noticed that when I've downloaded it in the past, whenever I am in Lightroom and I go to select Edit in Photoshop on an image that the only option I get is to edit it in PS Beta. I would have thought it would give me the option to edit in regular PS or in Beta, but it doesn't. The only way I can get around it is to open up PS first before selecting Edit in Photoshop and then it will ask me if I want to open it up regular PS or in Beta. Have you found a way to give you an option from the start rather than having to open up PS first? I've resorted to removing Beta because of this.
Thanks for your video Anthony - always very informative. The issue I have with this new AI feature is I fail to see how this is photography and will just further lead to more fake news. I appreciate that this has been possible for a long time through content aware etc but this just takes it to another level.
Wow, so finally I will be able to climb top of Mt. Everest in swimming suite!!
The layer has a layer mask. If something is deformed like the building in your video simply paint it back in from the original layer below.
does it matter if your using good graphic card or somewhat high end graphics card. i have geforce 3080 10gb and 32gb of ram. will that matter when using the ai program. i know it matters when i use tapaz photo ai since use my 3080 10gb and with my 32gb of ram on my desktop
You are using too simple prompts. You need to be more descriptive.
Hey Matt. Thanks for another great vid. Is there a way to export a photo from Lightroom to Photoshop beta ? Thanks
I'm wondering if you can select the mask segment of the top layer and mask out the newly generated bad roof, etc. ?????
Thanks for sharing. I've seen tons of glowing reviews. A good heads up that letting people know that if they have some issues with the Generative, it is probably not them.
One nice thing is if you wish to expand the canvas and then fill in.
Thanks for this. Question: When you generate from a resulting "generation" is Adobe regenerating from your original image or from the generated image?
I have used it and found like you that it has a long way to go yet. But it is quick compared to trying to remove people and rebuild the image.
Finally someone that shows the truth that it is not perfect. We need more of these real videos.
Hey Anthony: I downloaded the Bata version and found that when I selected edit in PS from Lightroom it opened the Bata version rather than PS 2023. Have you seen this happen ??? I ended up deleting PS Bata and it deleted the primary editor in Lighroom.
If you have the latest update to Lightroom Classic then in External Editing section of preferences there is an option to specify which version of Photoshop is the default if you have multiple versions loaded. Installing Beta makes it default so if you reinstall Beta and the go to Lightroom preferences you can then set default back to production version. If you want you can create an external editor preset for the Beta. I have no idea about whether there is similar option in Lightroom.
should i try some the beta shop and other beta stuff i own them.
Thanks for posting. You are correct. It needs a few iterations. The down side as you mentioned are the bad deeds which could come out of it.
Why would you want to isolate an out-of-focus background?
What happens when you want model to disappear if you type in the field , no person .
Yep, only Photog on earth not excited 😂😂
I used Generative fill to change an ugly bakcground. I must admit, the results are breath taking. I showed the pics to my model and she was confused. Huh.... We didn't go to beach, did we? I have no idea where this is all leading too but when properly used, it is certainly an usefull feature....
Finally. I’ve watched several CZcams videos on this from some pretty big tubers singing how awesome and “ game changing” it is. Adobe Photoshop’s biggest advance in decades. I’ve been experimenting with it and more often than not the results are pretty bad.
definitely hit or miss, I've gotten some things that look like cartoons. I entered a text prompt for "purple lightning" and it was just that, cartoon lightning
I don't like the direction we're headed with this. That being said, your expectations may be a bit high. You should also expand your selection so that it has room to play.
Having played with it for 30 minutes I find it very hit or miss and more miss than hit. However, if you want to add some background image (dog on street or boat in water) then it can be OK.
I imagine that other than one word in the prompt would work better? So, typing 'small red sports car', for example vs the SUVs you got :)
I viewed several CZcamsrs demonstrations of this Generative AI while waiting for yours Anthony as I have several concerns similar to yours. I'm a retired cop and I can see forensic photographic evidence being tampered with on both sides of the fence. Certainly there was a case a few years ago here in Western Australia where a cigarette packet was either added or removed from a scene. Scary stuff where you wonder about the integrity of the meta data...
Graphic design is an area where I can see numerous applications. I'm sure there'll be others but my mind is old and I'm not hip to all that's happening and the potential this tool offers YET. I have some ideas such as resizing images and others but I'll reserve judgement.
In an any event the tool bar I downloaded doesn't have the the graphics box to enter verbiage.
Thanks Anthony you're my Go To Man and my faith in you is confirmed.
Agree on the possible criminal use. Photographic evidence will be more and more problematic going forward. Even if Adobe ( and other) will have ways to prove that an image is a construct. Brave new world!!!!!
I briefly tried it. Didn't work that well. Maybe it still has to "learn" more lol.
The reliability of image creation is better with loose selections if you use tight " select Subject" expand the selection to improve results also mark the bad results with a thumbs down and good with thumbs up train the algorithm machine learning need the user to teach it
You are absolutely right, there are many who think it is a sensation, but unfortunately it has its limitations right now. I think it will take some time before we see it flawless, and it just needs some time.🙂
Aaron Nace (Phlearn) was pretty impressed, and it's doing better than he can with 15 years' experience.
It works a lot better if you use more than one word to describe the filled area. Just inputting "car" won't give you as good as a result as "car with open door."
The Contextual Task Bar doesnt show up
I've been very impressed with this. And this is only version 1.
The sick part of me kind of likes the unpredictability of the results
Hah! My first attempt was to remove/replace an ugly SUV that was spoiling any otherwise beautiful picture of a’63 ‘vette. It kept trying to replace it with an even uglier distorted truck. Gave up on that and tried “street”, “road”, and “building.” Still bad. I was about to give up when I randomly tried “brown bear”. Worked perfectly! I swear it’s impossible to tell the bear replaced an SUV. Looks totally natural!
For generative ai you don't have to accurately make your selection otherwise results will not be good
I've been trying to put an old fashioned tall ship on my picture of the chesapeake bay bridge and it won't do it. Sometimes it shows a Chinese junk. So..I tried to add birds to the picture of the chesapeake bay bridge. No matter how I described it, it gave me seagulls attacking the camera. First there was one. In another attempt it gave me two seagulls attacking the camera. Maybe I should use curse words, like "Give me the GD tall ship with sails unfurled, GD it!"
Actually they would canabaklize their sky-replacement-feature business, where they sell sky packages to customers. THhat would probably not run so well anymore if you could generate sky-replacements out of thin air.
Needs a lot of work, I find it great for extending photos changing background and out painting ie adding fish to sea composition etc but removing a whole subject always seems to leave deformed disfigured people. & PS has a perfect working sky replacement tool so when something works don’t fix it! So I agree definitely needs work before people start paying for credits, on top of the subscription once integrated into non beta PS.
It's in beta right now, it will get better. That being said, I think Anthony did a disservice by not using the tool properly. Consequently, the results are going to be skewed negatively.
The girl from your first image just appeared in my living room out of nowhere.
How do I send her back?
Why?
I am just uncomfortable with my photo being added to or modified by a machine. It's no longer my photo. If I add something that I have taken I am still uncomfortable - but less so!
The answer is simple - don't use it.
It seems to me the real skill required for any AI project will be in inserting good and relevant prompts
Cameras will be an optional, unnecessary, option very soon.
I think your selections are not suited to the task at hand, others have made simple loose selections, laso for example, with no refinement. I fixed reflections in my glasses with a (relatively) huge selection.
If I wanted the model getting out of a car, I would trn "getting out of a car." If you have me that instruction, I would do nothing, except perhaps object.
If you replace the sky, you need to fix the reflections. You didn't test that.
Respectfully I disagree. I think the creative applications of this are crazy & it will definitely change my entire creative process. Being able to not only re-image old work, but to actually see it? Nah that's crazy
I don't think you're using it properly, which may be why you're getting suboptimal results. How about looking at a few other YT photographers' channels to learn Anthony. I would suggest the excellent channel PTC Photoshop Training Channel to see where your selections could be improved. Of course it's not perfect. But you can get much closer by modifying how you're working. And this is just the beginning, Five, ten years from now? Scary.
Degenerative fill? ;)