Wow! This is so cool!

Amazing! It just keeps getting better. 💖 Thanks for the heads up!

Whoa. This is huge. I saw that there was 'symmetry' added but havent looked into what it is. Moebius I used for prompts at some point too, he signed a lot of stuff it seems! :) Everything kept coming out with attempts at his autograph :) This planet batch you start the video with is beautiful. Roger Dean and Moebius are a great combo here, wow.

new goodies! jumping in now. thanks for sharing this!

Fantastic videos and explanations, thank you very much ...keep going in this direction ;-)

Thanks for those videos, your channel is great!! 😁

I discovered these models last week. It's incredible! You are doing a great job with all these videos. I like the music also! Subscribed!

@QuickEyedSky

2 жыл бұрын

Thank you Maria! And indeed these are incredible tools... Focus on the prompt!

This is the shit I’m talking about. So excited to experiment with this

Stunning

As every time awesome

Cool,Thanks!

Thanks BRO

thannnnnk you

Super helpful videos. Thanks a lot. I wanted to get into the discord. Is it invite only? Cant get the link to work

hello I have a question can we run Disco Diffusion v5.3 on Paperspace Gradient platform instead of google collab thanks

Long live symmetry!

Hey mate, thanks for the video! what do you run DD with?!

@QuickEyedSky

2 жыл бұрын

With a web browser.

I wonder how this will fare with animation... Will have to try it today... haha. Great case studio... This can easily be achieved using Photoshop if people are using DD for images though

Wait a minute... this is epic and beyond words but... is that Old RS music I hear in the background?

@QuickEyedSky

2 жыл бұрын

Hi David, rarely people talk about my music... Yeah it's an old waltz 😌 I made decades ago. Vintage? I imagine I like to get out of the usual things. Thanks for noticing!

Quick question: I’m assuming yea, but the symmetry setting applies to animation frames, right?

@AZTECMAN

2 жыл бұрын

Animation frames are just using the init of the previous frame, where frame loss scale is exactly equivalent to init loss scale. The latest symmetry notebook uses a direct transformation (flipping each pixel) on a certain step of the iteration process (of a single image / frame). In other words, yes! Symmetry is completely independent of animation. Since each image is rendered separately for the animation process, a future version of DD could even include changing the amount of symmetry expressed on different animation frames!

I entered Colab to find that my 5.2 had been replaced with 5.3 so I did some experimenting with your prompt and mine. Had more pleasing results with horizontal symmetry than vertical and with both together, the result was awful! Possibly better used with fantasy rather than more photorealistic outputs.

@peterusher2020

2 жыл бұрын

An extra note: vertical symmetry does not give believable reflections, I have just discovered with a mountain, trees and lake prompt.

everytime i put a initial reference image . the session crashes showing cuda out of memeory . i use googlr collab pro + . waht can i do ?

@QuickEyedSky

2 жыл бұрын

Did you change the models? Isn't this image too big? Have you tried to disconnect and start from the beginning?

Same problem as DD always has: The images are always too bubbly and oily. I have tried and failed to get more orthogonal images from it. If you can get more angular more clearly defined pictures I'd like to see them! It's kinda like seeing shapes in bath foam and once I wanted something else more specific, well defined and angular I found I couldn't do it.

@pookienumnums

2 жыл бұрын

it 100% has to do with the prompt. its not a problem with disco diffusion. its a problem with you or your settings. you can get photorealistic imagery, angular shapes in a variety of mediums, and styles. i mean.. you can do shit that looks like basquiat if you wanted. try researching artists that dont look 'bubbly and oily' and try using them.

What are your settings? They look way more detailed than mine!

@QuickEyedSky

2 жыл бұрын

You can use an upscaler to get more pixels.

@jagshrapnel

2 жыл бұрын

@@QuickEyedSky I use Topaz, same?

My main complain with DD is that it takes about an hour for image, and it freezes regularly. Is there a way to get the images way faster and more reliable?

@ShawnFumo

2 жыл бұрын

One of the other videos that shows the difference between the tiers of Colab showed that speed was one of the big differences. So if you can afford to pay something, that should help for sure.

@mitchio83

2 жыл бұрын

Make smaller images and then use an AI upscaler is one option. But If I get a gpu other than the k80 i can do 1024x768 images in about 20 minutes. 512x512 in like 7 minutes. Colab pro is worth it though, its only $10 a month.

@pookienumnums

2 жыл бұрын

it largely matters on how many steps, your cut_n batches, and your cut ic power, and your cut scheduling. and the size of your image of course. 512x512 is NOT a bad starting point at all. you can even go smaller and the images still look good. then as the others said use an upscaler. colab pro and pro plus will enable you to get higher tiers of gpu to process through and that also absolutely has an effect. and, you can run dd locally if you have a good graphics card but i dknot know how to do that.

How does this performs with portraits?Thanks!

@QuickEyedSky

2 жыл бұрын

Hi. Not so bad. I'm testing...

@leunam3955

2 жыл бұрын

@@QuickEyedSky Thank you!

quelque chose me dit que tu pourrais être français...

@QuickEyedSky

2 жыл бұрын

Hé oui !