211–220 of 338 posts
Earlier quoted context omitted.
> (which the other's can't due to copyright issues) Be sure to pass this along to Microsoft's GitHub Copilot.
Nope copilot is only trained on open source codebases.
Earlier quoted context omitted.
The article isn't loading for me, so I can't really comment on the images it contains, but I've found telling the ai to apply an impressionistic filter does wonders for removing the unsettling aspect. Obviously that limits you to a specific style of image, but I imagine there are other stylistic filters you might apply that achieve the same goal. I could spend all day looking at the output of "impressionist cats" and…
I'm over 1000 credits into Dalle so far (I know, I know..) and you're on the money. You can go a lot further than impressionism, though. Specifying the names of famous illustrators, photographers or artists. Specifying the media used. Lens types. Colorways. Film types. Lighting. The right combinations can yield some incredibly realistic looking things, even faces, and then for the rest of it, there's Photoshop, Gigap…
Are training sets prepared with systematic variations in individual axes, as an alternative/addition to tagging each of millions of training images on these axes?
Earlier quoted context omitted.
Also there is nothing preventing Stock Image Sites themselves using Dall-E to generate additional images. Heck they can use their own existing images for training (which the other's can't due to copyright issues) to increase the portfolio, but the Stock Image Sites can access free public images. So, counter-intuitively it may strengthen Stock Image Sites value
Considering how DALL-E is fed with so many stock images that it sometimes spontaneously generates specific stock image websites' watermarks on output,[1] this is the stupidest, blandest possible ouroboros. [1] https://twitter.com/kevin2kelly/status/1551964984325812224
The same prompt with DALL-E gives me proper silhouettes with no watermark.
>"While the role of the artist isn’t going away soon, the role of stock image sites might disappear. " Not, yet. While it's cheap relative to stock images, it's time consuming to generate exactly what you want. Prices for stock images will collapse for the common quick to use images but the price for the specialized high end images will hold their value or even increase in value. Those historical and such images will…
Here's a vast catalog of Dall-E images and the prompts used to generate them. https://www.krea.ai/ Oh, and if you generate an image with Dall-E and there's a face that is distorted, you can use this tool to restore the facial features. https://arc.tencent.com/en/ai-demos/faceRestoration
Very interesting and promising, thanks! But I always get a 500 error when trying to upload an image... Were you able to make it work past the demo?
Earlier quoted context omitted.
> The images just need to get across a vibe, they don't need to be perfect I dislike super generic stock photo at the beginning of an article. It’s completely pointless, sometimes aesthetically unpleasant, often disconnected with the actual content, and hence a distraction. If neither you nor the reader cares about the stock photo, why not just forsake the thumbnail or use your website’s logo?
It seems pointless but it works. Our monkey brains like aesthetics. From the article: > Blog posts with images get 2.3x more engagement
Perhaps I’m just an outlier; but every time I see a thumbnail that has nothing to do with the title, I scroll past it. It’s a negative for me.
Am I one of the few people who finds these generated pictures really bad? They often have weird and unsettling details when you look closely. I mean, it’s an incredible achievement in AI that we can generate images at this level, but I don’t want them shown to me on a daily basis while I’m reading blogs.
I find the images to be incredible, but it's very unsettling when you focus on certain details like hands, feet, and eyes. The hands and feet that it draws are almost always mangled, and while it does a good job of drawing an individual eye, it doesn't seem to draw two eyes in a well coordinated manner, either one eye is bigger than the other, or there is something weirdly unsymmetrical about the eyes that makes the…
Earlier quoted context omitted.
Nope copilot is only trained on open source codebases.
open source isn't the same as public domain. the code is still copyrighted and has a specific license that has to be adhered to in order to be allowed to use it
So even if you are right, and courts rule that training is not fair use, it seems likely that all the big tech companies would lobby Congress to restore the prior state of affairs, as not having an ML industry is somewhat of a national security issue if the rest of the world is going full steam ahead on making Skynet.
Earlier quoted context omitted.
I find the images to be incredible, but it's very unsettling when you focus on certain details like hands, feet, and eyes. The hands and feet that it draws are almost always mangled, and while it does a good job of drawing an individual eye, it doesn't seem to draw two eyes in a well coordinated manner, either one eye is bigger than the other, or there is something weirdly unsymmetrical about the eyes that makes the…
I have a great example of this: "Jesus takes a selfie" - https://imgur.com/a/togE2ko I'm pretty sure this snap was three days after the resurrection.