Stable Diffusion XL 0.9 is here and currently available on Clipdrop! API access is coming soon along with a full model release in mid-july (we hope). This model is incredible, because it builds both on progress from developments in standard stable diffusion generative models AND deepfloyd! Fixing hands or mangled faces should be a thing of the past!
Let us know what you think in the comments!
PLEASE like and subscribe as well!
Announcement: https://stability.ai/blog/sdxl-09-stable-diffusion
Follow Emad on Twitter: https://twitter.com/EMostaque
Hello friends welcome to AI Flux stability AI finally released an update to stable diffusion XL they’re calling it version 0.9 and it’s kind of incredible so if you’ve been following a mod or stability AI on Twitter you’ve noticed that Emma has been teasing a lot of this model over the past few weeks the biggest one has been this sort of industrial area with a huge cloud of colored smoke but clearly used to build AI is working on really really big things what I will say is initially when I saw these I thought these were mostly coming from Deep Floyd because of how good certain textures were and predominantly how some of these had a lot of text in them there were also some that appeared to have a large contextual preference that seemed to favor deep Floyd overstable diffusion weirdly you know stable diffusion prior struggled with a certain context and certain things that could make some images harder to generate specifically this morning I was more so looking forward to fluid updates because the text here is good and and really the biggest uh breakthrough with deep Floyd was text but today we’re talking about stable diffusion XL so I really like how stability AI does announcements as an organization they’ve become really product focused and I think it’s helped them really sharpen why they’re making new models how they’re tuning new models and it’s also sort of changed their approach to safety they talk about safety way less unlike openai and talk way more about forward use cases of their systems so specifically in this case the new version of stable diffusion XL is available today in clip drop which is kind of the mid-journey interface-esque product that stability AI has now and they’re really pushing it so if you want to use this uh go make an account there you can use it today the API is coming soon and I’m sure eventually we’ll be able to get to as they say here a open source release in mid-july which is that’s where the safety aspect of this is coming in they say despite its ability to be run on Modern consumer gpus it’s able to Fusion XL 0.9 presents a leap and created use cases for generative AI imagery the ability to generate hyper realistic Creations for films these applications as well as offering advancements for design and Industrial use places sdxl at the Forefront of real world applications for AI imagery and they’re pushing it really really hard especially with clip drop and I think also clip drop is really one of the most well executed product deployments of generative AI for now at least in a real like kind of e-commerce and the number of other use cases they’re actually mostly comparing sdxl beta to their release today and the progress is actually pretty impressive so again the left is beta the right is the announcement today the biggest things I’ve been able to pull out of this are improvements in depth of field and color gamut and color gamut along with just how much color the model is comfortable expressing in any given generation so another thing that I think is cool is clearly they’ve trained this on some inputs from Fallout New Vegas because this Tower here is nearly identical to how the tower looks like in New Vegas clearly more context awareness of faces and organic shapes similar to hands so hands have also been vastly improved also curiously as a sort of a side effect of this text in stable diffusion XL is also much much better here on the left you can see this Fox’s face is kind of mangled and on the right this is nearing photorealism which previously was something that only mid-journey was getting close or was capable of doing this is a more minor uh comparison so we have a coffee here coffee here and what I think is telling again is depth of field a general contextual notion of subject and also understanding environmentally what looks a bit more normal the palm trees here are still a little wild but again the Right image is definitely more convincing if you show me the left image four months.
コメント