Photoshop adds AI tools to generate new layers

[Nikolai Fellini]

This quick video of the new AI tools in Photoshop is a game changer. For example: have a photo you want to add a dog to? Use the lasso tool to pick the area, type "dog" and it generates 3 different images of dogs. It can do all the things the AI image tools do, but it controls how & where you use it, in your images. It does not change the whole image, but creates new layers.

For example, with an SL pic, you wish the building in the pic was wider, lasso the area, hit "generate" and it gives you 3 versions of how the building would look, if wider.

I have never felt tempted to use AI in my images, but this is a big temptation. Who of us will move to the dark side?

 

 

 

[SLIM Barrowstone]

19 hours ago, Nikolai Fellini said:

This quick video of the new AI tools in Photoshop is a game changer. For example: have a photo you want to add a dog to? Use the lasso tool to pick the area, type "dog" and it generates 3 different images of dogs. It can do all the things the AI image tools do, but it controls how & where you use it, in your images. It does not change the whole image, but creates new layers.

For example, with an SL pic, you wish the building in the pic was wider, lasso the area, hit "generate" and it gives you 3 versions of how the building would look, if wider.

I have never felt tempted to use AI in my images, but this is a big temptation. Who of us will move to the dark side?

 

 

 

very nice i dont use ps but i will liik into this 

[ValKalAstra]

On 6/7/2023 at 5:18 PM, Nikolai Fellini said:

Who of us will move to the dark side?

Honest answer: Most already have, it's just they didn't necessarily realise it. Image editing programs/apps for the last decade or so have been making use of machine learning to tweak existing tools and new ones (E.g. those creepy face app filters that were all the rage for a while). It's just that it has now become a hot button topic through copious amounts of FUD. Personally, I've been on the dark side for a while, whatever that means to me it means learning. I'd rather get ahead of technology, understand it and learn about it, instead of fearing it. Am I super comfy about all the implications? Hell no. But knowledge triumphs over fear.

That said, Inpaint/Outpaint (what Photoshop is doing there) is quite impressive as demonstrated. Something else that has impressed me was Segment Anything (https://segment-anything.com/demo#), combined with Grounding Dino, it allows you to designate areas of a picture and (using grounding dino) describe them and then have the program create accurate masks. I just mention this because i know I am not the only one that hates masking.

The next impressive tech is DragGAN, where you can basically drag a 2D picture in a 3D way. The demo video shows what it does (https://github.com/OpenGVLab/InternGPT/tree/main#draggan_demo) That one doesn't seem consumer ready yet by virtue of needing excessive hardware power. For now. Other interesting applications are things such as T2I style transfer or using control nets with img2img in Stable Diffusion to modify the lighting in interesting ways.

 

 

 

[Carpentoon]

This is amazing!

[Rosemary McDonnell]

It's not just PS Beta. It's the AI/ML masking that has become available in Lightroom Classic over the last six months. Masks for body parts are pretty close and only need a bit of work if that. Adobe is also trying to monetize AI/ML work so if part of your image in the Adobe Cloud is used in another image you get a small residual. BTW, PS Beta seems to be getting at least one update a week now.