I just posted about this today in another thread. The ability for AI to be able to describe photos for people who are visually impaired is a total game changer. Most visually impaired people rely on people posting AltText or photo descriptions to become involved, but as most of us already know, most people don't think twice about it, including myself. I'm definitely trying to be better.

I've used it a bit to try and work on my Spanish. That is, using it as a sophisticated chatbot. Unfortunately it's still quite frustrating for that: I figured I'd ask it to play un juego de rol (a roleplaying game), and it kinda sucks at it. I'm gonna give it a go with an open source alternative, hopefully they're less aggressively calibrated toward being tedious and awful. It's just, getting an open source language model running takes a decent amount of time and effort, so I'm sorta midway through that.

I was doing some photography with a 50mm equivalent prime and didn't want to go back to my car to get a wider lens for a shot that I was a bit too close to get the framing I wanted for. So I took 6 photos of the scene and stitched them together as a panorama with Hugin.

Turns out I was missing the bottom right corner and I had a big black square ruining the shot. So I took the post-processed panorama into GIMP, drew a crude outline of what I thought would be there and used StableDiffusion's Img2Img to generate the missing bit.

A little messy, but better than losing the image entirely.
(Unfortunately the image includes things that would get me doxxed instantly)

As for LLM's. Apparently GIMP doesn't have a bulk export function that exports all active images at once. So instead of exporting 140 images that I had masked for photogrammetry manually one at a time, I asked Bing Chat if it could make me a GimpFu script to batch export.

It gave me this:

from gimpfu import *

def export_all_images():
    for image in gimp.image_list():
        filename = image.filename
        if filename:
            new_filename = filename.rsplit(".", 1)[0] + ".jpg"
            pdb.gimp_file_save(image, image.active_layer, new_filename, new_filename)
        else:
            print("Image has no filename. Skipping.")

export_all_images()

which worked and saved me quite a bit of time.

One cool thing is to summarize YouTube videos and ask it to ignore all the fluff.