Adobe Photoshop's New 'Generative Fill' AI Tool Lets You Manipulate Photos With Text (arstechnica.com) 38
Adobe has introduced a new tool called "Generative Fill" in the Photoshop beta, which utilizes cloud-based image synthesis and AI-generated content to fill selected areas of an image based on a text description. Ars Technica reports: Powered by Adobe Firefly, Generative Fill works similarly to a technique called "inpainting" used in DALL-E and Stable Diffusion releases since last year. At the core of Generative Fill is Adobe Firefly, which is Adobe's custom image-synthesis model. As a deep learning AI model, Firefly has been trained on millions of images in Adobe's stock library to associate certain imagery with text descriptions of them. Now part of Photoshop, people can type in what they want to see (i.e., "a clown on a computer monitor"), and Firefly will synthesize several options for the user to choose from. Generative Fill uses a well-known AI technique called "inpainting" to create a context-aware generation that can seamlessly blend synthesized imagery into an existing image.
To use Generative Fill, users select an area of an existing image they want to modify. After selecting it, a "Contextual Task Bar" pops up that allows users to type in a description of what they want to see generated in the selected area. Photoshop sends this data to Adobe's servers for processing, then returns results in the app. After generating, the user has the option to select between several options of generations or to create more options to browse through. When used, the Generative Fill tool creates a new "Generative Layer," allowing for non-destructive alterations of image content, such as additions, extensions, or removals, driven by these text prompts. It automatically adjusts to the perspective, lighting, and style of the selected image.
To use Generative Fill, users select an area of an existing image they want to modify. After selecting it, a "Contextual Task Bar" pops up that allows users to type in a description of what they want to see generated in the selected area. Photoshop sends this data to Adobe's servers for processing, then returns results in the app. After generating, the user has the option to select between several options of generations or to create more options to browse through. When used, the Generative Fill tool creates a new "Generative Layer," allowing for non-destructive alterations of image content, such as additions, extensions, or removals, driven by these text prompts. It automatically adjusts to the perspective, lighting, and style of the selected image.
Re: (Score:3, Funny)
Seems this NaziGPT needs lots of work.
Re: (Score:2)
Seems this NaziGPT needs lots of work.
They can't even drive through a gate [cnn.com]. What makes you think they could form a coherent thought?
Re: (Score:1)
It's Gategate, bigger than Watergate. We're on to them this time!
Re: (Score:3, Insightful)
Why don't you post about everything the White House and thus CNN censor like a transsexual shooting people in a church
You mean the story CNN covered [cnn.com]?
Lately, one side has become more dangerous than the other
Yes, one side [yahoo.com] is decidedly more dangerous than the other, especially when they call for the killing [newsweek.com] of people they don't like, support terrorists, and defend child rapists [imgur.com]. You were saying?
Re: (Score:2)
I'm still VERY curious as to why the transgender terrorist you are talking about , that shot up the church, has not had her manifesto published yet?!?!?
Seems that would have been released long ago...what's holding that up?
Re:flynn's digital warriors fucking the world (Score:1)
For April Fools' Day... (Score:3)
For April Fools' Day, could they release a version that interprets the prompts as James Fridman would, doing something that is technically correct, but not remotely what any reasonable person would likely have intended?
Re: (Score:2)
They could call it Genie Mode.
Who among us hasn't wasted an afternoon with D&D players trying to craft a genie-proof wish?
Midjourney, openart.ai can do that too (Score:2)
probably better and cheaper too.
Re: (Score:1)
Is it integrated into an image editing tool masked and layered to specific part of an image? No? Then it can't do shit. No one here is claiming that Adobe built something that is better than Midjourney or Openart.ai. They are claiming such a tool has been integrated in a useful workflow.
Please learn what is being discussed before commenting nonsense.
HookGPT (Score:1)
> which utilizes cloud-based image synthesis
Adobe is trying to get you addicted to their fee-cloud by making fancy options that depend on cloud.
Re: (Score:3)
> which utilizes cloud-based image synthesis
Adobe is trying to get you addicted to their fee-cloud by making fancy options that depend on cloud.
Adobe basically stopped selling software a while back and started renting it, so everything Adobe makes depends on the cloud. Now they're just trying to get back a few of the myriad people who gave them the middle finger because of it.
Re: (Score:3)
Re: (Score:3)
So the two software companies who have most seriously abused their market positions over the last two or three decades both stopped doing perpetual licenses? I'm shocked. Shocked! Well, not that shocked.
Re: (Score:2)
I don't see myself going back to Adobe in any shape, form or fashion any time soon.
I don't rent my software.
I've found software just as good if not better in:
Affinity Photo
Affinity Designer
Affinity Publisher
Capture One
ON1 RAW
Davinici Resolve (and the included tools fo
Re: (Score:1)
There's lots of hacked-for-desktop copies of Photoshop floating around.
Re: (Score:2)
As a legitimate owner of Photoshop CS6 and Lightroom CS6, which fail on modern versions of macOS only because of defective copy protection code, and otherwise work correctly (based on the fact that it works exactly once after migrating to new hardware), let me just say that the temptation to hack Adobe's software has never been greater.
Creative Cloud is an abomination, and I'm now wary of perpetual licenses from companies that also offer subscriptions, out of fear that they'll also pull an Adobe.
Re: HookGPT (Score:2)
Re: (Score:1)
It's not a matter of IF, it's a matter of how.
Subject to their approval (Score:3)
Interesting, since I don't remember Photoshop ever having built-in automated censorship before. Why not? It would be easy enough for the Text tool to have a built-in black list of forbidden words.
Re: (Score:3)
Not that I'm currently a paying customer, but if I were an artist there is no way I would pay for a tool with built in censorship.
Sex and violence are part of life (enjoyably and regrettably in that order), and thus part of art. Who would buy a tool that might not let them work on a nude, or a scene involving implied or explicit violence?
Worse, "confidential"? Who the hell knows what that means other than sometimes the tool is going to refuse to work because you MIGHT be offending Disney or something.
Re: (Score:2)
Not that I'm currently a paying customer, but if I were an artist there is no way I would pay for a tool with built in censorship.
Pretty much every single commercial image editing suite since the late 90s / early 2000s will refuse to scan, import, or print accurate images of banknotes from ~60 different countries: https://en.wikipedia.org/wiki/... [wikipedia.org]
Re: (Score:2)
...which is terrible because there are plenty of non-illicit reasons to do that. It's perfectly legal to reproduce the currency of many nations at other scales, or in other colors, or if you only do them one-sided, or some combination of these— so long as your intent is non-criminal. e.g. for US currency [uscurrency.gov] the reproduction may be in color or black and white, but must be under 75% or over 125% size and one-sided.
In case you're wondering what they do in movies, like I was, they alter the money to make it
Re: (Score:2)
Not that I'm currently a paying customer, but if I were an artist there is no way I would pay for a tool with built in censorship.
While I agree with the principle of what you're saying, if you're in the industry that uses Photoshop you'll pay for photoshop despite this and despite your moral objection. The reality is Photoshop is leaps and bounds ahead of any competition for serious (reads: I make money doing this) image editing. It's a largely hated tool that people can't live without.
Maybe that will change. We used to say similar things about Premier before people started jumping ship to DaVinci Resolve, but right now there's no too
Re: (Score:2)
Long ago I moved to Affinity Photo [serif.com] from Photoshop...and have never looked back.
The engine in AP is new from the ground up and much faster than old
Re: (Score:2)
Try Photoshopping a dollar bill or other bank note and it will stop you in your tracks.
Re: (Score:2)
Photoshop has had built in censorship since the late 90s at least. Try scanning a bank note at high resolution and see what happens.
https://creativemarket.com/blo... [creativemarket.com]
Mixed Bag (Score:3, Informative)
I am subscribed to Adobe Photo Cloud, so I am able to try the beta...
I took a picture of me and my wife at Machu Picchu (so lots of old rocky strictures and grass/trees), and tried a few enhancements...
1) Add Llama. Dropped in a Llama with a very realistic looking shadow, but the Llama itself looked just like the kind of vaguely Llama shaped thing you expect from AI.
2) Tried "remove people" circling a couple standing behind us next to a rock wall. This did a passable job, the rock wall we were standing by continued onward with the people gone.
3) Tried "remove people" after circling a stairway and ledge off in the distance that had various people on it (so people maybe 20 pixels high or so). Here the AI went a little hard, and while it did remove the people it took out the whole stairway and also add a big tree I didn't care for. Looked through the three options and one of them kind of had a mushy looking stairway... I wouldn't want to use any of the three options it had unless you were just wanting a small print where I guess it might pass.
At least each of the Generative Fill passes comes in as a layer so you can quickly toggle on or off to examine and discard.
Followup (Score:2)
After looking at the output again, the AI really messes with stuff..
The rock wall I mentioned that it generated to replace people, was a totally different kind of rock wall with very large rocks instead of a smoothed surface.
I was so busy being creeped out by the Boston Dynamics-esque Llama it made, I didn't notice it took out a whole building behind the Llama and replaced it with a rock-fenced grassy courtyard!
It also replaced parts of a mountain in the background with a different, yet similar mountain fac
Re: (Score:2)
It doesn't have to be perfect to be good enough for mass producing fake social media posts in the run-up to an election.
Llama says maybe not (Score:1)
It doesn't have to be perfect to be good enough for mass producing fake social media posts in the run-up to an election.
Maybe, but if generated people re altered people are as messed up as this Llama, people will notice... :-)
Or in Free/OSS land (Score:5, Informative)
There's a Krita plugin for stable diffusion...
Changing aspect ratios on older TV shows? (Score:2)