Tag: stability.ai

Stable Diffusion XL Beta on the AI Horde!

Db0

In the past month or so, I’ve been collaborating with stability.ai to help them improve the quality of the new upcoming SDXL model by providing expertise around automation and efficiency in quickly testing new iterations of their model using the the AI Horde as middleware.

Today I’m proud to announce that we have arranged with stability.ai that the new SDXL beta will become available on the AI Horde directly, in a form that will be used to gather data for further improvements of the model!

You will notice that a new model is available on the AI horde: SDXL_beta::stability.ai#6901. This is the SDXL running on compute from stability.ai directly. Anyone with an account on the AI Horde can now opt to use this model! However it works a bit differently then usual.

First of all, this model will always return 2 images, regardless of how many you requested. The reason being is that each image is created using a different inference pipeline and we want you to tell us which you think is best! As a result, the images you will create with this model are always shared! This means that the images you generate will be stored and then used for further improvement based on the ratings you provide.

Which ratings? Well each time you generate images with SDXL you can immediately rate them for a hefty kudos rebate! Assuming your client supports it, you should always report back which image is better than the two. You can then optionally also rate each of them for aesthetic ratings (how much you subjectively like them) and for artifacts (how ruined the image is from generation imperfections like multiple fingers etc)

Lucid Creations has already been adjusted to support this!

For best results with SDXL, you should generate images of 1024×1024 as that is its native resolution. The minimum allowed resolution for SDXL has therefore been adjusted to this.

Also, as I mentioned before, this beta is primarily going to be used to improve the model, therefore we’re disallowing anonymous usage on this model for the moment. However registering an account is trivial and free and will also give you more priority in your generations!

The most cool factor is that as stability.ai further improves the model, is will automatically become part of this test, without you having to do anything! As such, you will eventually start getting better and better generations from the SDXL_beta on the AI horde. This is why you continued sharing of the quality is very important!

So please go forth and use the new model and let us know how it performs!

Stable Horde receives stability.ai processing power!

Db0

A week ago I mentioned that we had begun a collaboration with LAION to provide them with ratings on images. The amount of ratings we have received since then has blown away all our expectations! In just a week, you’ve all rated close to 130.000 individual images! As a comparison, the LAION-aesthetics v2, which was instrumental for training Stable Diffusion v1.x, used less than 600K rated images. We’ve reached 1/4 of that amount in a week!

Needless to say, these amounts seemed to turn some heads to the power of mutual aid provided by the stable horde, and some gears were set in motion.

LAION spoke with stability.ai directly and arranged that it would likewise benefit them to support the health of the stable horde itself. Since stability.ai is set to be the most direct beneficiaries of a better trained the laion-aesthetics v3 it makes perfect sense.

I was not privy to the discussions that happened, but I was happy to learn that Tom, the CTO of stability.ai arranged to provide us with some sponsored resources in the form of 4 VMs with RTX4000s Nvidia GPUs!

Quite surprisingly I had to deploy the VMs myself, so I crafted the most optimal setup for taking advantage of those 8Gb of VRAM through my experience with my own RTX2070. Each of them has been loaded with standard stable_diffusion 1.5 and 2.1 and each of them then has 8-10 other finetuned models to help cover the versatility provided by the Stable Horde. Granted, we are serving close to 100 different models currently, but the fact that those workers will remain running consistently 24/7, should help provide cover and allow other workers to switch to less supported models as well.

I hope this is the start of a fruitful collaboration between the stability.ai and the Stable Horde. The way I see it, the current scenario is a win-win for everyone. We get a more consistent service which allows more people to use it and makes them more likely to rate images to give back, which are then fed back to LAION and by extension stability.ai.