# Stable Diffusion Version 2: Power To The People… For Free!

## Метаданные

- **Канал:** Two Minute Papers
- **YouTube:** https://www.youtube.com/watch?v=HytucGhwTRs
- **Дата:** 11.12.2022
- **Длительность:** 7:21
- **Просмотры:** 210,663

## Описание

❤️ Check out Anyscale and try it for free here: https://www.anyscale.com/papers

Stable Diffusion version 2 release notes:
https://stability.ai/blog/stable-diffusion-v2-release

Try it on the web - https://huggingface.co/spaces/stabilityai/stable-diffusion
Or run it locally: https://github.com/Stability-AI/stablediffusion

Refractive images - @recatm - https://twitter.com/recatm/status/1596933520672583680
Textures - @recatm - https://twitter.com/recatm/status/1596933527836119040
Humans - @EMostaque - https://twitter.com/EMostaque/status/1596620680442703873
Cyberpunk book cover @technollama - https://twitter.com/technollama/status/1597219897683378177
Interiors - https://twitter.com/williamcusick/status/1597022736957591553

Stable Diffusion Web for searching prompts: https://stablediffusionweb.com/prompts
Luxcore render: https://luxcorerender.org/

Interpolation: https://twitter.com/xsteenbrugge/status/1558508866463219712
Full video of interpolation: https://www.youtube.com/watch?v=Bo3VZCjDhGI

🙏 We would like to thank our generous Patreon supporters who make Two Minute Papers possible:
Aleksandr Mashrabov, Alex Balfanz, Alex Haro, Andrew Melnychuk, Benji Rabhan, Bryan Learn, B Shang, Christian Ahlin, Edward Unthank, Eric Martel, Geronimo Moralez, Gordon Child, Jace O'Brien, Jack Lukic, John Le, Jonas, Jonathan, Kenneth Davis, Klaus Busse, Kyle Davis, Lorin Atzberger, Lukas Biewald, Luke Dominique Warner, Matthew Allen Fisher, Matthew Valle, Michael Albrecht, Michael Tedder, Nevin Spoljaric, Nikhil Velpanur, Owen Campbell-Moore, Owen Skarpness, Rajarshi Nigam, Ramsey Elbasheer, Richard Sundvall, Steef, Taras Bobrovytsky, Ted Johnson, Thomas Krcmar, Timothy Sum Hon Mun, Torsten Reil, Tybie Fitzhugh, Ueli Gallizzi.
If you wish to appear here or pick up other perks, click here: https://www.patreon.com/TwoMinutePapers

Thumbnail background image credit: https://twitter.com/recatm/status/1596933520672583680
Thumbnail background design: Felícia Zsolnai-Fehér - http://felicia.hu

Chapters:
0:00 What is Stable Diffusion?
0:27 Higher resolution images
0:59 Depth+text to image
1:53 Easier inpainting
2:17 Reflections, refraction
2:46 Photorealistic humans!
3:25 Interiors
4:05 Textures
4:22 Try it yourself!
5:04 Free and open source
5:37 What about making videos?

Károly Zsolnai-Fehér's links:
Instagram: https://www.instagram.com/twominutepapers/
Twitter: https://twitter.com/twominutepapers
Web: https://cg.tuwien.ac.at/~zsolnai/

## Содержание

### [0:00](https://www.youtube.com/watch?v=HytucGhwTRs) What is Stable Diffusion?

Dear Fellow Scholars, this is Two Minute  Papers with Dr. Károly Zsolnai-Fehér. Oh my, here is Stable Diffusion version  2. So, what is this? Stable Diffusion is   a free and open source text to image AI,  which means that we write a piece of text,   and it creates exactly that image  for us. All this, for everyone,   for free! So good. And now, here are 10  things you should know about version 2.

### [0:27](https://www.youtube.com/watch?v=HytucGhwTRs&t=27s) Higher resolution images

One, it can now generate images with higher  resolution. More details for free, that sounds   amazing, but, it doesn’t stop there, because two,  it can also perform super resolution better. Super   resolution means that in goes a coarse image,  and out comes a beautiful image with a lot more   detail. And when I say a lot, I mean a lot.   Just look at how much better this is. So good. Three, it can also go from depth plus  text to image. What does that mean? Well,

### [0:59](https://www.youtube.com/watch?v=HytucGhwTRs&t=59s) Depth+text to image

we can give it an input image, and then,  it fires up this paper. Oh goodness,   that is a powerful paper, it can not only  estimate depth maps for photos really well,   but get this, it can even imagine if a drawing  were a real photo, and estimate the depth of   that. Incredible. So, it creates the depth  map for this image, and using our prompt,   fills it with information that makes sense. This  is excellent if we wish to just specify the pose   of our hero, and it will generate as many  similar variations for it as we can imagine. Or, we can also do this. Lots of amazing variants   for the same concept. The limit is  only our imagination. Loving it.

### [1:53](https://www.youtube.com/watch?v=HytucGhwTRs&t=113s) Easier inpainting

Four, image inpainting is now  easier. This means that we can say   that we wish to retain this part of the  image, and the AI should just fill in the   rest according to our instructions. This is  super fun, and it will help with one of the   most important things we could all ask for: and  that is quick turnaround times for our ideas.

### [2:17](https://www.youtube.com/watch?v=HytucGhwTRs&t=137s) Reflections, refraction

Five, it can now generate more convincing  images of refractive objects. You know that   I am a light transport researcher  by trade and this makes me very,   very happy. Human eyes, and water, underwater  images, not a problem. And some of these   images are extremely convincing. These  are especially impressive because they   came about from pure prompting, no  guide or any input images were used. Six, now hold on to your papers for photorealistic  humans. Whoa, these are shockingly good.

### [2:46](https://www.youtube.com/watch?v=HytucGhwTRs&t=166s) Photorealistic humans!

Especially that we are also humans, most of us  anyway, and we really have a keen eye for other   humans, and if even the smallest things are  off, we notice immediately. And the details   in this version are finally at the point where  it can create believable virtual characters. But if photorealism is not your cup of tea, and  you like virtual worlds better, do not despair,   seven it is also excellent at  creating cyberpunk book covers.

### [3:25](https://www.youtube.com/watch?v=HytucGhwTRs&t=205s) Interiors

Eight, we can also generate incredible interiors  with it. Now, not all of them are perfect,   there are some flaws here, but the  pace of progress in AI research is   nothing short of amazing. Not so long  ago, we needed proper light simulation   programs and hours of work to  create an interior like this. And now,   just a text prompt. 5 seconds. And just imagine  what this will look like just two more papers   down the line. What a time to be alive! Note  that I put a link to the authors of these   images in the video description so make sure to  check them out if you wish to see more of these.

### [4:05](https://www.youtube.com/watch?v=HytucGhwTRs&t=245s) Textures

Nine, the textures are also incredible.   Here, you don’t see super fancy images,   just simple concepts that are executed  really well. I just kept on looking and   looking at each of these images  and I can barely find any flaws.

### [4:22](https://www.youtube.com/watch?v=HytucGhwTRs&t=262s) Try it yourself!

Ten, of course, I can only imagine how  excited you Fellow Scholars are to try it,   so, can you? Yes, and it gets better, because  you can still try it in two different ways. One, if you are patient, you can write a prompt  here. You might have to wait for a bit, but as   of the making of this video, it works. Now what  happens when you Fellow Scholars get over there,   who really knows, we have crashed plenty of  websites before with our Scholarly Stampede. But,   if you don’t want to wait or run  some more advanced experiments,   you can run the model yourself at home  on a consumer graphics card. Loving it.

### [5:04](https://www.youtube.com/watch?v=HytucGhwTRs&t=304s) Free and open source

And don’t forget, Stable Diffusion is free  and open-source, and thus it has captured the   imagination of you Fellow Scholars. With this, you  can pop the hood, take out your virtual wrenches,   and, let the experiments begin! And with this,  AI-based image generation is only getting   cheaper and more democratized from here on out.   A little open-source competition for OpenAI and   Google. Power to the people, and for free.   Double thumbs up! What a time to be alive! By the way, Google already has an AI that creates  not only images, but videos from your prompts,

### [5:37](https://www.youtube.com/watch?v=HytucGhwTRs&t=337s) What about making videos?

and if everything goes well, I may or may not have  exclusive access to it. If it comes to fruition,   there will be a video on it with my  own Scholarly Prompts, so consider   subscribing and hitting the bell icon, you  really don’t want to miss it if it comes. Thanks for watching and for your generous  support, and I'll see you next time!

---
*Источник: https://ekstraktznaniy.ru/video/13361*