SDXL 0.9 Rivals Midjourney v5.2, You can create 12000 Images for FREE, Right Now!
Things are getting crazier folks. Stability AI has released a near completed version of their XL project, as SDXL 0.9. The missing 0.1 means that some kinks are being worked out & SDXL 1.0 will be released very soon.
Read on to find out more about all of this.
The most wonderful part about Stability Ais Stable Diffusion is that it is completely open source. You can create unlimited images from this tool & if you want to create your app like Stable Diffusion webui then you can implement Stable Diffusion 0.9 from Hugging Face.
You will find all necessary resource links at the bottom of this article.
SDXL 0.9 Examples
Just look at these examples & tell me if they don’t look amazing. Considering that Midjourney charges $8/200 images/month & we can generate 12000 images/month for absolutely FREE, the ball is definitely in SDXL 0.9’s court.
SDXL 0.9 – SDXL 1.0 – System Requirements
You will be deeply glad to know that SDXL 0.9 doesn’t require a beefy system. I hope you already know that text to image, image to image & such types of AI based generation requires a good amount of GPU VRAM & a powerful GPU specs-wise.
The lowest you can go with SDXL 0.9 is just 16 GB of RAM & 8GB of VRAM. If you have just a little bit of mid-end gaming computer, then you can run this model beautifully.
Of course, with better specs, you’ll be able to generate images faster.
But just knowing that SDXL 0.9 is going to be compatible with nearly all mid-end gaming systems is amazing. Kudos to the developers who kept the optimizations top notch that everyone can benefit from this model and generate their own images without paying hefty monthly subscription fees to third party apps.
Why is SDXL 0.9 so amazing?
Stability AI has beefed up the parameters fed to the model. So, SDXL 0.9 was trained with 3.5B parameters for Base Model & a whopping 6.6B parameters ensemble pipeline.
According to Stability AI, the final output will be delivered after the first model create the entire image & then the second model refines it to a degree that feels almost magical.
The max resolution that it currently outputs is 1024×1024, which you can then scale-up using several scaling models.
But just the fact that beautiful, hyper realistic & true-to-prompt images will be generated in 1024×1024 resolution, its just a cherry on top of an already pretty sweet cake.
SDXL’s Beta was quite successful
The beta figures were in favour of Stability AI’s SDXL. It was launched on 13th April 2023 & nearly 7000 discord members generated an staggering 700,000 images.
Several images went into the showdown challenge & nearly 3500+ images were chosen as winners.
That figure alone is astonishing given that several other contenders also submitted their respective images.
SDXL 0.9 / 1.0 Features
The model already supports better resolution & better understanding of the prompt to generate the images.
Apart from that it supports in-painting & out-painting. We are already quite familiar with in-painting but I am curiously looking forward to out-painting.
It is being said that out-painting has been improved a lot. Now, you just either provide an image or a prompt to let SDXL 0.9 know what needs to be added to the image.
This would help a lot when building a landscape around an object. The applications are huge & I do hope that these features keep evolving because I can imagine that it would be great for developing realistic worlds for games quicker than before.
SDXL 0.9 / 1.0 Release Date – Availability
https://clipdrop.co/stable-diffusionI got a little late for the party because SDXL 0.9 is already out. You can take a look at the implementation at Clipdrop app by Stability AI (links are at the bottom).
Paid subscribers of Stability AI API & DreamStudio also have access to this version, since 26th June 2023.
NightCafe app will also provide an option to choose SDXL 0.9 to generate images.
Right now, developers cannot access the model directly. If you visit the resources section of this article, you’ll find a hugging face link.
The link is currently accepting applications only. So, once your application is approved, you will be able to get your hands on this model to build your own app around it.
As for SDXL 1.0 release date, it is being mentioned as mid-July, which may or may not get delayed. I’ll keep you posted about it.
SDXL 0.9 License – Can you use it commercially?
Sadly, SDXL 0.9 is currently available with a non-commercial license. It means that you can use it for research purposes only.
I am sure that once SDXL 1.0 comes out, we’ll have more commercial control of the model. But again, I can’t comment on that in advance.
Now you must be wondering…
What is Google doing? What is Google’s Generative AI Studio?
If you are then check out this article on the same – Google Generative AI Studio – Will it ever launch?
Don’t Miss the Video Below
If you want to checkout what kind of images SDXL 0.9 can generate, then checkout the video below. It showcases the capabilities quite nicely.
You can also checkout the second video that compares SDXL 0.9 directly with Midjourney.
Pingback: [N] SDXL 0.9 Released – SDXL 1.0 Release Date – How to Access SD XL Right Now? – Code Jana – Machine Learning and Artificial Intelligence