Dramatically BETTER RESULTS in Generative Fill, Photoshop TIP
HTML-код
- Опубликовано: 29 май 2023
- 2 powerful tip on getting much better results with Adobe Generative Fill in Photoshop. Colin Smith shows you the unexpected reason your results are bad in Photoshop Generative Fill
► My new course on Generative Fill in Photoshop: photoshopcafe.com/shop/genera...
Save 20% with code GFYT20
► Free Photoshop add ons: photoshopcafe.com/vault
► I'm a BenQ Ambassador and use their monitors: www.benq.com/en-us/campaign/d...
► THE GEAR I USE: www.bhphotovideo.com/c/browse...
► www.amazon.com/shop/photoshop...
► THE MUSIC I USE: share.epidemicsound.com/photos...
► SUBSCRIBE FOR MORE VIDS: ruclips.net/user/photosho...
► PREMIUM COURSES: photoshopCAFE.com/video
► INSTAGRAM: / photoshopcafe
► WEB: photoshopcafe.com
► FACEBOOK: / photoshopcafe
► TWITTER: / photoshopcafe
#Photoshop #adobefirefly #generativefill
As usual... a great tip and soooo well explained. Excellent work, Colin.... 🙂
I noticed that the start of a selection will generate the direction of an object. If you make a selection for the cat from the other direction, it will be laying in the other direction.
Great tip
Wow thank you
Interesting.
As always, you rock Colin. Thank you for the very insightful, useful practical tips!
Awesome tips! Thank you so much Colin!
terrific tip. Appreciate your continuing to share your Photoshop expertise.
Thank you and yes it was very informative and I was not aware that the shape matters in generative fill. Thank you again.
Really great video. Thank you!
Very helpful!
Great tip. Thanks Colin
Thanks Colin. That is a very useful tip!
Many thanks again. Use your tips and advice a lot.
Very useful important method! Would like to see more on how you can combine other Photoshop techniques with Generative fill when it doesn't get things quite right.
I've been using Midjourney for several months but was trying to get Photoshop to behave like it. Your tip really makes sense and is another dimension not available in MJ . BRAVO.
Yeah, prompting in MJ is quite different in Ps
That's very helpful, thank you!
Totally makes sense. Thanks for the tip.
Awesome!! Thank you for confirming these.
Very enlightening! Thank you.
I ve been playing whit this for hours, and this video manage to provide me an extra tip. Great job.
Super helpful, thank you
That was great, thanks for the video!
EXCELLENT VIDEO!!! THANKS FOR SHARING!!!!! 👏👏👏
Very helpful...thank you very much
Very interesting. Didn't know. Thanks!
thanks for the tip!!
very helpful tnx
Great tip about the shape of the target fill.
Shape as well as the size is important. I had an image where the hand was very blurred and I wanted to replace it with a hand where you can see the knuckles and the wrinkles. I outlined the hat but left some gap around the edges, and the result was a much larger hand. If you want to generate something that is precise, the size and accuracy of the outline is extremely important.
My new course on Generative Fill in Photoshop: photoshopcafe.com/shop/generative-fill-in-photoshop-introduction-course/
Save 20% with code GFYT20
Biggest thing to ever happen in Photoshop - Generative Fill: ruclips.net/video/VEWTJcpENu0/видео.html
Expanding the edges of a photo: ruclips.net/video/se4V-PyO2Rg/видео.html
Generative Fill livestream in depth!: ruclips.net/user/livewsXnPamM6uo?feature=share
Thankyou. I had worked out that it was important to make the selection big enough but I had not noticed that the shape was also important. By the way, you don't make it clear here but it is also often important to overlap your selection slightly with the part you are not changing so the AI knows it needs to make the new part match up with the existing part eg to make sure that cat looks like it is sitting on the speaker rather than just being above the speaker for example, it is allowed to make the paws extend over the front.
Thanks for this tip. Extremly important.
Useful. I kinda figured that the shape would be important!
Thank you for your great tip
Ohhhhh!! Thankyouuuuuu! ♥
Very useful - I am learning all my Generative Fill from you - thanks so much!
I'm glad you are enjoying it!
I cannot stand the guideline violation errors. Restrictive and unpredictable.
Interesting Tips , and good to know, Thanks.
Brilliant Colin. Thank you.
Thanks
Excellent tip regarding the shape, have not seen this anywhere else. Thanks.
You "probably" will now lol. I have been ripped a lot over the years :)
Great tip! 👍🏾🙏🏾
Wow so good!
Fantastic tips🎉
Great video!!
Thank you so much for this. I was starting to get a sense that my selection was important but didn't know it was crucial to help PS determine what to generate.
Excellent as usual.
Another clear explanation Colin 👌
Thx
that helps thanks
Amazing, fabulous tip!
Thanks
Thanks
Thank you! I was just messing up with Photoshop (beta) and this video just got posted and answered my question. What is the odd?
of course! good shapes arre always works better even in content-aware too
You should do one on dealing with the user guideline violations. It is very prohibitive.
Very useful
Great thank you , now my fill will be good.
Another winner, Colin.
Bobble Hat in the UK Colin, nice vid 😉
Really great
Colin, that is a very useful tip! I don't know what 'pompom' violates, but I got that when on a photo I took of a fun fair being dismantled with no people in the shot that I'd split toned and made very gritty, I wanted to make it even more Fallout/post apocalyptic wasteland by adding a child's "broken doll" - nope. That violates. It's not like I asked for "bIow up doII", that could have been wrong on 2 counts.
I have noticed that AI understands words even if it is spelt incorrectly or missing letters. I think in this case 'pom' was recognised as 'porn'.
@@jacekkomeda7087 To be fair, once A.i. can do THAT, it will no doubt skyrocket like home video and the internet did.
That first tip was definitely the piece i was missing. Great fallout
Ok, now I know why I wasn’t getting good results! 😂
Programmers have a word for the brevity and clarity of your explanations: elegant!
Thanks so much!
Bobby hat in UK. I keep getting bizarre messages about violating the rules even when it's produced dozens of images using the same prompt! Thanks for the useful video
as usual great tutorial. In England we called them a "Bobble hat" I will try and let you know
good to know
I shoot imteriors. I would like to select kitchen cabinet surfaces and replace with some of my own surface images. Can’t seem to figure out how to make it use my images
So, should we start saving selections? I do a lot of architectural renderings, so chair and couch-shaped selections; door, and window selections; Chandelier and other lighting?
Not a bad idea
So I made a sort of tall selection around two brown, deadish looking palm trees in an image of mine from San Diego, and typed "Two palm trees against a blue sky with light clouds," hoping to get some new, green palm trees. I got several generations of freeway or parking lot lights. One had a cell tower. I took out the "sky and clouds " part and just just typed "palm trees." I got ONE great looking palm tree...
P.S. Growing up in Minnesota we called them "stocking caps." I still do.
Try doing it step by step. First select the old trees and hit generative fill to remove them. Get the result you want and flatten the image / merge the layers. Then select the sky and ask for blue sky with light clouds. Again, get the result you want and flatten it. Finally, select the area you want the trees and ask for palm tree or two palm trees, and see how you go!
do you know how to upscale the generative fill element with something like gigapixel ai or such
I'm having trouble getting original hands to stay in tact on dancers and trying to put in a different ai generative fill background. Any suggestions?
Can you make a video on how to avoid user guideline error message? I get that a lot on simple non violation things.
This is helpful! When I try to add animals, it frequently suggests toy animals. Eg, I put "sheep" into generative fill, and it gave me a zombie sheep, a weird bad 3d render sheep, and a toy figurine of a sheep. No actual sheep. I have a similar result with prompts like "cat," they are often weird 3d render versions of the animal. Is there any way to get it to provide more realistic animals rather than cartoon, toy, or 3d render animals?
Wonderful tip. Thanks! Speaking of violating user guidelines, I'm getting weird rejections. For instance, I was trying to remove salt and pepper shakers from a tabletop, and it wouldn't do it due to a guideline problem. This sort of thing is happening often. Any thoughts?
I think it says that when it cannot work out how to do it.
@@RobinHartJones Could be. I hadn't thought of that.
The user guideline violation messages are baffling sometimes. I’ve got them when trying to remove an object in front of some bushes with no prompts from me and in other weird use cases.
size too for perspective
In Canada we call those hats Toques. I cannot WAIT to check out your new course on PS - AI!!!
Good to know. Thanks, the course is coming along great!
Alright you win 👏🏽
Didn't you do the pompon hat demo yesterday? Incidentally, I fired up the beta and confirmed that you get a hat (and no guidelines warning) if you correctly type “pompon” (decorative puffball on a garment) instead of “pom pom” (which also can be a military slang term for a rapid-fire antiaircraft gun; the Associated Press stylebook used to distinguish between the two, and I'll bet Adobe followed AP style when keywording their training set.) The guidelines generally don't like prompts involving weapons; either you get the warning or a completely different object of the same general shape.
And I never knew it was pompon. You learn something new every day :) thanks
As a non native english speaker I've learned a ton of words since I've started to experiment with generative AI.
@@photoshopcafe In the UK, I have always known them to be called bobble hats. The 'bobble' being the bit on the top. Thank you for another great and informative video.
Another fun instance of the no-weapons rule: yesterday I wanted to depict a guy wearing one of those short fleece-lined jackets that we in the US call a "bomber jacket." But when I typed "bomber jacket" into the prompt box, I got the content-guideline warning. So I typed "flight jacket" instead and got the garment I wanted.
Ive probably watched every video out there on generative fill. You are the FIRST to discover the importance of the shape..
Generator fill working in offline?
Woollen hat with ball thingy is called a toque .
Nice
I get very frustrated by the violation warning, as I am usually just asking to remove a hanging light, a soda can, etc. I hope Adobe fixes this.
All I get are error prompts 😢... I'll have to wait for an updated release I guess.
As a fellow Kiwi, ask it to fill with a New Zealand flag the results are quite funny. I also get very strange violations. My best tip is alway keep the hands, face and feet out of the selection and as you have shown in another video do it in parts for the best results.
Lol oh boy, a Union Jack and a made up flag
Valuable content. Well done!
Thanks
Bobble Hat in the UK
bro apply your dob option is not coming when i login to behance
Try it in CC
I notice that in the first hour testing.
Colin, can this generative AI be used to add someone from another photo in to another one proficiently? If so, how?
Not really with any quality right now.
2good sir
It’s interesting how you can control the results of the AI based off the shape of your selection. Very useful tip.
By the way, do you know if generative fill uses the same technology as the remove tool when removing objects? I know they both use AI to remove objects, but is the only difference that one is selection based and the other brush based? It might be interesting to see a comparison between the two methods. Like pick an object to remove with generative fill, then do the same thing with the remove tool and compare the results. As of right now it seems the remove tool is faster; however, you don’t get variations/options to choose from like you do with generative fill. But it would be interesting to see a comparison in the quality they produce. Anyways thanks for all you videos!
They are both powered by Adobe sensei, so I would imagine they are similar in many ways, but possibly trained different? That’s a good question. I’ll see if I can talk to Adobe and find out
Remove tool is using a lightweighted AI model so it can run on-device. So you won't need to upload the image to the server and everything is processed in your machine. It supports any resolution and preserve the high quality textures. The backends of the remove tool is GAN-based so it does not work well with novel object completion. Like if you want to remove a person in front of another person, it will fail. However, the GenFill is powered by diffusion-based models so it is more powerful for generating novel scenes. But you have to upload the image to the server for processing, and it can only support 1K resolution so the textures won't be good enough.
@@yuqianzhou4646 thank you for that information it is very helpful!
@@yuqianzhou4646 great response!
So weird having Tommy Lee Jones giving me Photoshop tips, Amazing thank you!!!
Boom!
How is your generative fill prompt always showing?
When you have a selection active, it appears. You can also open it from the menu Window>Contextual Task Bar
Great, useful tips! "Pom pom" has a definition on Urban Dictionary that might explain why Adobe didn't accept it.
There's a lot more to Generative Fill than I suspect most of us initially realized. Between yourself and Jesus Ramirez - the other channels tend to put up similar content a day or two after yourself and Jesus, whatever that means 😉 - we are learning lots of cool tips and tricks for getting the most out of it. I'm actually a lot happier with Generative Fill and Firefly after watching the PS Cafe and PS Training Channel, because you've demonstrated that there is still skill involved in getting the best results. What has surprised me is that I have already used those tools in projects, mainly for creating backgrounds, but I got nearer to the wrinkled fisherman I wanted with Firefly than I could find from stock images. BTW Does your bobble hat comment mean that you were originally from New Zealand? You have an unusual accent, that I can't place.
Yes, I’m originally from NZ (and Scotland). Jesus and I seem to be a source of “inspiration”
I'd love to see a video on having AI work with an image of an object that I've created. For example, let's say I have a photo that I took off a custom pencil and I want Photoshop to put it into a goose's beak. Is this possible? Thx!
Nobody is talking about where do the AI objects come from. Adobe Photo Stock? What about rights in commercial use?
Not sure where you have been? Everyone is talking about it. Watch my first vid on this series.
THE TED LASSO TOOL!!! lololol
👍👍
I can't seem to generate people. The people that Photoshop generates look like zombies or something straight from a horror show. Can you do a video on how to get better results with humans?
Right now, it’s hit and miss with people. It will get better
I guess if you see a photo with a cat in an image that you like, you can copy the selection of it, put that selection in your photo then add a cat that is positioned similarly.