These have both been taken with the exact same camera from the same location. The one on the left is with the OnePlus camera app, and the one on the right is from a community modification of the Google camera app to work on the OnePlus 12. The Google one looks a lot better because they use super-resolution from multiple short exposures automatically.

The Google camera app does not usually look better without zoom (in my short time testing) and also has a harder time focusing.

With the smaller 14b model (q4_k_m), just letting it complete the text starting with "why do I"

edit: bonus, completely nonsensical (?) starting with "I don't" (what could possibly be causing it to say this?)

I was thinking about how hard it is to accurately determine whether a screenshot posted online is real or not. I'm thinking there could be an option in the browser to take a "secure screenshot", which would tag the screenshot with the date, url, and whether the page was modified on your computer. It could then hash both the tag and the image data and automatically upload this hash to some secure server somehow. There would need to be a way to guarantee that only the browser could do this, or at least some way to tell exactly what the source was. I'm not much of a cryptography person, but I would be surprised if it isn't possible to do this. Then, you could check if the screenshot you see is legitimate by seeing if it's hash exists in the list of real hashes.

reference image if you have no idea what I'm talking about:

I know this is a minor nitpick, but it's something that annoys me.

I got this graphics card mostly because it was the best deal on Amazon at the time (gpu shortage), and I also thought it looked decent from the images they had. However, when I actually installed it, all I see is the relatively unattractive looking black metal backplate with some white text. The other side is always the side shown in the promotional images too - not a single one of the pictures in the Amazon listing even shows the side that you'll be seeing 99.9% of the time. Do they think everyone hangs their PCs above them from the ceiling, or has open-air testbenches? Why do they never even bother with the other side? I know they want the fans on the bottom so the cooling is better, but the air in front of the CPU shouldn't be that bad, a lot of cheaper GPUs don't need that much cooling, and a ton of people have watercooling now anyways so the CPU radiators just go on the sides.

my reasoning: the actual colors we can see -> the wavelengths that we can extrapolate to -> basically extrapolated wavelengths plus an 'unpure-ness' factor -> not even real wavelengths (ok well king blue and maybe lavender if I'm being generous could be)

Just 3% less votes than Jill Stein, and he dropped out 3 months ago

This is at JFK, does anyone know what they are used for? There wasn’t an obvious time when it was taking a picture.

Prompt: A cyberpunk scifi painting of a floating city in the air above the sea

It uses a new, fancier, 18GB text encoder (t5) to follow the prompt much more closely. It isn't perfect, but its much better than SDXL in my opinion. It does seem to be a bit worse at photorealistic subjects and has a tendency to create 1-pixel vertical lines.

Some other images:

impressionist, a woman sits in the middle of a crowded cyberpunk street, people bustling around, orange and blue glowing signs, warm atmosphere

a bright cinematic photo of a solarpunk city at midday, skyscrapers, steel, glass, vines and fields of vivid tropical plants

I get around 1 image every quarter of a second on my 3060. The quality isn't up to par with regular SDXL (not even close) but it follows prompts well and is extremely fast. Here are some of the best images in this batch:

Prompt: "impressionist oil painting, watercolor, a crying old southern man eats cheese at sunset in front of a futuristic dystopian cyberpunk city"

Material: 3D model: Original image: