Cool Pictures from the Mind of a Machine: AI Generated Pictures

[Gedemon]

locally, yes, ~5 minutes for a 49 frames video, IIRC those GIF above are at 16 fps

 

[Samson]

AI generated pop video

Spoiler More videos :

 

[Bonyduck Campersang]

AI generated pop video

Spoiler More videos :

I guess this is what it would look like to watch your dreams

 

[Bonyduck Campersang]

Face in the Mist

 

[Bonyduck Campersang]

Bing AI

 

[Gedemon]

generated locally

SDXL + CogVideo

 

[Birdjaguar]

The grey one is mean looking!

 

[Gedemon]

The grey one is mean looking!

yes, some models can give human expressions to a lot of non-human characters.

 

[Gedemon]

a few wallpapers generated locally, using some illustrious models (anime trained SDXL) for the base composition with a latent upscale using more photorealistic models to add details.

 

[Birdjaguar]

How about a nice beach (sand and waves) desktop?

 

[Gedemon]

How about a nice beach (sand and waves) desktop?

drinks included

 

[Birdjaguar]

The drink and bar spoil it. Too much rock. Needs a bit more wave action.

 

[Gedemon]

I'd have to change the workflow for that, I'll keep the request in mind for when I'll try something else.

 

[Gedemon]

to explain a bit, the anime model I use in the first pass to get the composition/colors/lighting of that series is not good at realistic waves, and I'm not happy with the results in my few tries. And if I switch to a realistic model, I can get waves, but I lose everything else, and the generated picture would be in a very different style of the previous ones.

 

[Thorgalaeg]

For the looks I bet you used a Flux based model for the realistic pass. With all the custom models we have, we can do anything combining models. That was the midjourney way I suppose. You can also do an inpaint with other sdxl model in the sea to get birdjaguar's waves and then render it all with flux to get a cohesive image.

Btw is midjourney still a thing?

 

[Gedemon]

It was a realistic SDXL model for the second pass, I don't have enough VRAM to do a 2560x1440 upscale with flux (or I'd need to use Ultimate SD Upscale which I'm not good at for details using something else than SD 1.5)

Here is one with Flux

and an image to video generation with CogVideo

 

[Birdjaguar]

AI does cats pretty well.

 

[tjs282]

The image generators do have a lot of source material for that particular subject, after all.

You've even helped them yourself

 

[Samson]

TIL that image generators can make QR codes pretty

 

[Samson]

AI does cats pretty well.

Do you think? I think that looks a lot less like a real picture of a cat that say the pope image that started this thread. I am not exactly sure why, but even in hollywood films animals seem much less believable in CGI that humans.