Generating Photorealistic Images With OpenAI GLIDE

GLIDE is OpenAI's new Computer Vision (CV) model that performs better than DALLE. I show you how to generate images with GLIDE in this video. The images it generates are very high quality, some are nearly photorealistic. You can also make AI generated art with the model just by changing the prompt! The model is 3 billion parameters, though the public model is only 300 million parameters.

[mikeanima]

Maybe I'm a geek but this will never cease to be the coolest thing ever to me. It's the fact it's not frankenstein-ing existing images together, it's creating them from scratch, pixel by pixel

[OmegaKillerArtMeisterSkywalker]

Picture number five looks really cool! Definitely has a mysterious AI vibe to it.

[Katy-fvoz]

It's sad they don't give us the full model, imagine the fun that can be had with it.

[pavelkalugin]

I think the fifth picture looks great, something like tree traversal with a demonstration of combinatorial explosion

[sveinndagur]

They are so evil for not allowing us to make porn with this.

[vojtechhavranek]

Easy choice. Futuristic clipart robot, please!

[maxchronos]

Right now, I'm testing the limits of its power by writing "A monster dragon in a clown suit"

[EdLrandom]

yo yo yo, here from coffeezilla collab, great job!

[dank]

Amazing video!!

Would have been good to constraint the prompts to be pixel art to get more visually pleasing images.

You could also apply style transfer to edit them instead of photoshop 👀

[jettube]

I like profile pics 1, 4, and 5 personally. 3 is also good but I'm not sure if it fits.

[RaviAnnaswamy]

thanks for the great introduction and demo. I think if you are expecting a stick figure of a man etc you should prompt as 'a cartoon stick figure'
right now, it is interpreting is probably as figure of a stick.

[janislionel]

Would it be possible for you to provide your adjusted version? It's really nice that you can save to drive and have the prompt as a file name. Can't figure out how to do this myself.

[ONDANOTA]

I used it. I typed "A lion on an ice cube" and waited.... it's awful. It's like a totally flat black non realistic lion in front of a low quality ice-ish wall.
The other prompts did not go well either, except "A Maserati car" which is meh (number plate 100% wrong, but I did not expect it to be able to read (or write, in this case). Anyway, this is a good starting point

[JamesPound]

Is it possible to run the full this using the GLIDE model? The results are already amazing.

[erickwalterbutler]

cool vid, is it possible to use this to generate larger images than 256x256?

[zan_drox]

Great video! Link to the file you modified for batches?

[FranAbenza]

Obviously your channel is going to grow. You gotta upgrade the design of it before this explodes. But none of the generated images is truly future proof. If any, n1 and stick figure better represent your content the rest are kinda random :)

Keep up the good work

[real-prizrak]

Bro, awesome video, which is that Distro or OS? Looks so smooth

[rolininthemud]

I like #4 It looks like a cloud with a little face

[drdca]

pictures 3 and 5 look the best out of the 5 to me.
The robot face one (17:28) could also be good. maybe.
The one that looks kind of like an apple / restaurant logo is also amusing, not sure how well it works though.