I've been studying Stable diffusion for some days and this Github page I linked in the title shows the incredible power that Stable Diffusion has Right now, and it is growing everyday. I think people on HN ought to know and discuss about it, and share their ideas and discoveries.<p>It can do text-to-image out of nothing with a customer-grade GPU, and image-to-image for incredible variations that you can guide with the criteria you choose.<p><a href="https://imgur.com/a/Jt76AHA" rel="nofollow">https://imgur.com/a/Jt76AHA</a><p>Just check that picture out, I made variations from the base picture top-left, either to black-ened, or to white-ened, or to the combination of both... The power is infinite. It is called "Prompts matrix".<p>You can run it from your own computer preferably with 8GB VRAM+ GPU, or cloud-based virtual machine (I managed to run it on Microsoft Azure but not for cheap). To install Stable Diffusion and experiment for yourself, simply follow this page guidance: <a href="https://rentry.org/voldy#-guide-" rel="nofollow">https://rentry.org/voldy#-guide-</a><p>If you just want to try out simply you can check out "beta.dreamstudio.ai" but it's not free beyond the first 200 images. It is free on Huggingface but it's struggling with the demand so I Won't link it.<p>It generates 512x512 imges, but you can ENHANCE the fuck out of it with this link: <a href="https://github.com/xinntao/Real-ESRGAN" rel="nofollow">https://github.com/xinntao/Real-ESRGAN</a>
Check out the link "Portable Windows". You can commandline it and batch a whole folder with a .bat script, on Windows (and Unix of course). It's free, no strings attached (as far as I know).<p>If you generate texture or paintings you can generate a normal map and a displacement map, with CrazyBump.com. You can BATCH it a whole folder to their respective normals. There is a free trial which does the job, but buy it if you can, because it's awesome.<p>Here is a cool link about SD: <a href="https://lexica.art/" rel="nofollow">https://lexica.art/</a> ("the SD search engine")
Here's to showcase the most important PROMPTs the engine can handle: <a href="http://stable-diffusion-guide.s3-website-us-west-2.amazonaws.com/v1/people.html" rel="nofollow">http://stable-diffusion-guide.s3-website-us-west-2.amazonaws...</a>
At this link, you can buy and sell PROMPTS: <a href="https://promptbase.com/" rel="nofollow">https://promptbase.com/</a> (it's a great place to have great ideas)
Finally here's a database of prompts and their generated picture, to get ideas: <a href="https://laion-aesthetic.datasette.io/laion-aesthetic-6pls/images?_sort=rowid" rel="nofollow">https://laion-aesthetic.datasette.io/laion-aesthetic-6pls/im...</a><p>That's it for what I learned as of know. I hope people will talk about that and so we can get more nerds into the Stable Diffusion and make it even more open and awesome.<p>Thank you for reading