AI software can be run locally, i.e., entirely on your PC/laptop, using online servers, or using a combo of both. AI runs best on GPUs [graphics processors], which work differently than CPUs -- CPUs can work, but AI running purely on a CPU [or CPUs] is Extremely slow. AI also favors GPUs with LOTS of dedicated RAM. And AI works with some GPUs MUCH better than others... at the consumer level, GPUs built into the CPU will almost never work. The new Intel and Nvidia graphics cards are designed to work better with AI than earlier models, though those will usually work, and the same is supposed to be true for AMD's upcoming cards. More expensive cards are more powerful with generally more RAM and work better with AI than the cheaper models. While the NPUs [Neural Processing Unit] in Copilot+ laptops are designed for AI, they're generally too anemic to be much use, at least the current generation. Microsoft has tailored a few AI features in Windows that work exclusively with those NPUs, ignoring much more powerful GPUs when present, but then that was part of the marketing campaign that Microsoft hoped would make them money from new laptop sales. Note that AMD, Intel, & Nvidia GPUs each have their favored AI frameworks, so some AI models will work better than others &/or require modifications depending on the brand of GPU.
While most photo-centric AI software either uses online servers running the AI or integrates the models it uses with the software, SoftOrbits AI Photo Editor 10 comes closer to running a complete AI model locally. Per the in-app notice the model requires 12 GB of video card RAM, and without that downloaded model the AI portion of the app's GUI was grayed out. It *looked* like it just added the same sort of text generation that Microsoft's added to Copilot, Photos, & Paint, where you describe what you want & the AI generates the image. Otherwise the app is just another photo editor. Besides the app & the optional AI model it installs an older copy of the Python coding language/runtime. It *may* be able to work with the latest version if it's installed afterward, but I didn't try. The software is installed to Users\ [UserName]\ AppData\ Roaming\, but can be moved afterward. That does however trigger an error that a text file has the incorrect path, that appeared to be based on a hard-coded value [changing paths in the registry made no difference]. That said, the error did not seem to effect anything. The Python installation accounted for most of the new registry entries -- Regshot2 recorded 675, including entries just from Windows itself running -- and adds 20 files to Windows\ Installer\ plus py.exe, pyshellext.amd64.dll, & pyw.exe were added to Windows\.
What can AI do for you?
As everyone knows by now, AI can generate convincing-looking photos from a text prompt, but AFAIK it has not reached the point where you can click a button and get a fully edited, optimized photo anywhere near the quality that you can achieve on your own. It can make an image appear a bit sharper, but it cannot actually sharpen a soft or out of focus photo. AI can find the edges of objects, e.g., people etc. in a photo, but generally not with the best accuracy -- you can get away with it for a generic online photo, but for pro-level work you have to tweak the results. A couple few AI apps can do a great job reducing noise, it can work very well for enlarging/reducing the size of a photo, and sometimes it enhances the results of the preset looks that you can find in several apps nowadays. AI can excel at removing or replacing things in a photo, or not -- results really vary depending on the photo, what you're trying to remove or replace, and generally it seems, on the AI's mood. [Seriously, it seems like it will work great one day, but the next day something Very similar will fail completely.] When it does work it's very nice because it can match the texture or grain and lighting.
For photo editing, the most common use for AI is selecting subjects or objects in a photo, where AI finds the edges of the object so you can change either what's inside or outside of that boundary. With Windows Photos or an app like Aiarty Image Matting you can change the background for example. In Photoshop it lets you select a group of objects, like a group of people, or each individual, or parts of each individual, like their shirt, or face, or any skin showing, or eyes, or mouth, or lips, or teeth, all pretty much automatically. You can select the shirt, and change it into a different kind of shirt, or a jacket if you'd rather, and of course in whatever color you choose. The similar but somewhat reduced feature Photoshop Express lets you replace the eyes. A commercial artist might use the built-in image generation to alter a stock photo to met their design requirements. When you use the AI image generation you can have the generated object look like a reference photo to various degrees, and it will be placed in the photo where you want at the size you want with matching lighting etc. In Photoshop AI can also work to extend a photo, generating the necessary background to for example change a 4:3 photo into 16:9 -- depending on the background it can work surprisingly well. Depending on the photo, AI can work somewhat to colorize B&W photos -- most often you'll have to finish the job manually.
What AI cannot do is fix the colors. It won't remove a color cast or tint, it can't adjust colors that are off, and it cannot adjust the levels any better than previous, old fashioned auto levels adjustments, and you can still do better by hand. [You want the darkest part of the photo to approach pure black, the brightest pure white, and the mid range light enough to see detail without looking washed out.] AI can smooth a face, but it cannot edit out a blemish or wrinkle like you can using frequency separation. It can sometimes brighten teeth or improve eyes, and might be able to change eye color, but it's not reliable for any of that. Photoshop's AI Neural Filters can do some interesting, and sometimes surprisingly good quality editing and FX for faces, including aging and the reverse, or turning a frown into a smile, though again, it may or may not work, and you'll often have to do additional work to fix it or blend the results.
In summery, I'd rather have the AI in Photoshop than not, though most of the time it does not have a huge effect on the results. I do always use AI noise removal [from Topaz], but then I'm working with old analog photos [film]. Whenever I come across a new AI-based app or tool, e.g. SoftOrbits AI Photo Editor & Aiarty, I try it out, but so far haven't found anything I felt was worth pursuing -- I couldn't or can't see how it might improve the photo editing I do now. That's not to say I'll quit checking out new software or versions of software, and it's not to say you shouldn't try out any of this stuff for yourself. This is just my experience in case it helps.