# Stable Diffusion XL Is Here!

## Метаданные

- **Канал:** Two Minute Papers
- **YouTube:** https://www.youtube.com/watch?v=kkYaikeLJdc
- **Дата:** 11.08.2023
- **Длительность:** 6:04
- **Просмотры:** 112,915
- **Источник:** https://ekstraktznaniy.ru/video/13080

## Описание

❤️ Check out Lambda here and sign up for their GPU Cloud: https://lambdalabs.com/papers

Try it here online: https://clipdrop.co/stable-diffusion
How to run it: https://stable-diffusion-art.com/sdxl-model/
Run on all platforms: https://github.com/AUTOMATIC1111/stable-diffusion-webui
Alternative, modular solution: https://github.com/comfyanonymous/ComfyUI
Great Mac app: https://drawthings.ai/

Danielle Baskin: https://twitter.com/djbaskin/status/1514735924826963981 
SDXL Doré drawings: https://www.reddit.com/r/StableDiffusion/comments/155bwgz/gustave_dor%C3%A9_drawings_sdxl/
Orcton: https://twitter.com/OrctonAI/status/1684344552654610434

My latest paper on simulations that look almost like reality is available for free here:
https://rdcu.be/cWPfD 

Or this is the orig. Nature Physics link with clickable citations:
https://www.nature.com/articles/s41567-022-01788-5

🙏 We would like to thank our generous Patreon supporters who make Two Minute Papers possible:
Aleksandr Mashrabov, Alex Ba

## Транскрипт

### Segment 1 (00:00 - 05:00) []

Dear Fellow Scholars, this is Two Minute  Papers with Dr. Károly Zsolnai-Fehér. Today we are going to see some amazing, and some  less amazing results from Stable Diffusion XL,   the popular text to image AI’s new version  that you can also run for free online,   or even at home. So, what’s new here? Well, one it offers higher resolution  images and is better at challenging   concepts that previous text to  image AIs faltered at. Human hands,   very specific spatial arrangements where we  say that a woman should be in the background   chasing a dog in the foreground. These now  work better. However, I’ll immediately say   that this does not mean perfect. Here are some  of the results I generated locally. Obviously,   it cannot be anything else but images of Scholars  holding on to their papers, or at least studying a   huge magical tome. It can do that, however,  look. Hands still seem to be an issue. If you have a favorite artist, and you love  the style, but would like to imagine what if   he explored painting different subjects, you  can now do it at home, for free. I think this   is an amazing tool to explore new artistic ideas,  and I have to say it is incredibly fun too. And,   when comparing to Midjourney, I think the  quality of the results are better there,   however, SDXL seems to be more true to the  original style of the artist. Loving it. I could also not resist trying Danielle  Baskin’s amazing drink prompts,   which are also working quite well. Yummy. Now, they say that users generally  prefer the new technique’s results   to previous versions of Stable Diffusion,  although I haven’t seen this user study   being linked to a peer-reviewed paper,  so a wise Fellow Scholar like you are   does not take these results for granted.   We are going to try some more ourselves. Two, simpler prompting. If you have played  with previous versions of Stable Diffusion,   you probably noticed that it requires a very  detailed description of an image to create   something decent. But here, it is easier to  create something with just a few words. My   experiments have mostly matched this. When asking  for illustration of a small modern house in Osaka,   I got quite usable images that I liked.   This layered cake in the style of a   landscape was also just a couple words, and  the results are incredible. This is so much   fun! I hope you Fellow Scholars will also  come up with some crazy ideas using this. Three, it finally supports better text generation.   Now, writing text is typically quite difficult   for text to image AIs. When I asked for the full  “Two Minute Papers” in writing, I did get better   results than with most previous techniques  I’ve used, but most of the time, this was a   little too much to ask for. After many tries,  I had some success with it. Kind of. However,   I found that when asking for just “SDXL”,  even then it took a bit of trial end error,   but I finally started seeing some more success.   We are not there yet, but this is just the 1. 0   version of Stable Diffusion XL, so I am  excited to see how it improves over time. Also do you remember ControlNet? ControlNet is  a neural network structure that helps us give it   additional inputs beyond just text to image. And it is absolutely amazing. For instance,   we can provide just the edges of an input  image. A rough sketch, or edges extracted   from a real photo, and we get a beautiful image  with exactly that creature and framing. Other   kinds of inputs also work with ControlNet,  and as the preliminary results show, this   feature is coming soon to Stable Diffusion XL,  which will bump up its usability significantly. And we get all of these for free, forever. That  is excellent. It is all so new and so fresh,   there aren’t too many results out there  yet. And don’t forget, there are many   ways to improve this base model through  checkpoints and magical things called LoRAs,   so specialized versions of SDXL  will come out as soon as weeks,   or perhaps days from now. It will be so  much better. What a time to be alive! And if you wish to try it in your browser  or run it locally, I got you covered with   some links in the video description.   So for now, let the experiments begin!

### Segment 2 (05:00 - 06:00) [5:00]

Thanks for watching and for your generous  support, and I'll see you next time!