Automatically Detecting and splitting recognizabl objects into layers

Oooh... I just saw a video about one of the upcoming Adobe MAX Sneaks called Project Stardust that lets you move content around from a flattened image as if it were on a layer:

• https://fb.watch/ns9t1iTTBh/

And here's more info:

• https://labs.adobe.com/projects/stardust/

If you're making a feature request for Adobe's AI to unflatten a layer into layers, then you might try posting it to the Firefly Community:

• https://community.adobe.com/t5/adobe-firefly/ct-p/ct-adobe-firefly

Please show what you are talking about – all the images related to a meaningful example and the intended result. 

To be brunt:

This is a Forum for Photoshop users, so you will likely not get a lot of useful advice on training an AI model of your own. 

And whether the AI-related features implemented in Photoshop so far would be useful depends on what you are actually talking about. 

If the content was a Smart Object that contained layers, then you could use Layer > Smart Objects > Convert to Layers or in the Layers panel right-click on the layer and select Convert to Layers.

However, if you're starting with a flattened image with a single raster layer, then you could try using Illustrator's Image Trace (Window > Image Trace...) which will convert the raster content to vectors. You would need to expand (Object > Expand...) to modify the traced content. Then you could try scripting to separate the vector content onto separate layers.

In the most recent version of Illustrator, you can try Retype (Beta) (go to Type > Retype) to convert the vector tracing of the rasterized text to editable text. Here's more information:

• https://helpx.adobe.com/illustrator/using/retype.html

There is a premium third-party plugin for Illustrator from Astute Graphics that also converts text to editable type.