Reply To: AI Lets get real

SteveM CompChair

    I have been using Leonardo.AI for the past few weeks, and I’m trying hard not to be seduced by AI.

    I’ve gotten deep enough into it to see trends and possibilities, and as Competition Chair, I see a clear context where AI can be used and still fit within our competition rules even though it amplifies one’s abilities greatly.

    A feature of Leonardo.AI is the ability to create what is known as a custom AI model by inputting one’s own images to train it (I made a custom model using images from my portfolio entered in 2021 – see below). Let’s say that I import 500 of my own images to train the model. The textual prompt that I might use to instruct the processing would go something like this:

    “Using only images within my model and no other image source, extract and isolate interesting objects from my images and create a fictitious scenario that layers these objects into a realistic arrangement and landscape where these objects appear in a realistic context as well as make an intriguing scene. Use cinematic lighting with a 50mm perspective and output at a 4×1 aspect ratio.” …you get the idea. The prompt can be much more specific such as specifying the location: at Ocean Beach in San Francisco just past sunset with large waves crashing on the shore. This assumes that I have one or more images of this scene in the model trained from my own images and that there is sufficient imagery contained therein for the AI to identify SF’s Ocean Beach.

    It can generate up to eight possibilities at a time.

    It will then do in seconds what would take ages: go through 500 images and manually remove backgrounds from objects and save the objects in separate files or layers to be composited together and then play with dozens and dozens of clipped image objects to arrange into a realistic imagined scene.

    One can already import one’s own images and have it creatively fix or change part of the image, but one can’t limit it 100% to one’s own images (SRPS / N4C rules), but I’m sure that capability is not far in the future.

    Our rules rely on the honesty and integrity of the member. AI can’t change that.   😉

    Here’s one of the input images:

    An Illustrated Man-AF Layers v2

    And here is one example of many it generated: