ATK
New Delhi [India], December 9: Stability AI, the company behind Stable Diffusion recently launched their latest Stable Diffusion model called SDXL Turbo. This model brings real-time image generation which changes how artists use Stable Diffusion to generate images.
Unlike other Stable Diffusion models that require multiple steps for image generation, SDXL Turbo is a single-step image generation model. This speeds up image generation times dramatically making real-time image generation possible.
This development comes as a surprise for many users, especially those using tools like MidJourney or Dall-E.
Stability AI has always been at the forefront of generative AI with Stable Diffusion and all its models. Due to its free and open-source nature, Stable Diffusion has advanced quite a lot through community-trained models.
Currently, AI images generated using Stable Diffusion represent 80 per cent of AI images on the Internet. With users generating over 2 million images every day using Stable Diffusion, this open-source project has been growing rapidly.
With the introduction of SDXL Turbo, the future of AI image generation looks even more promising as users can now generate images as they’re writing their prompts.
How Does SDXL Turbo Work?
SDXL Turbo uses a new ADD technique that allows single-step image generation. The model is trained on images of 512×512 dimensions.
Stability AI has launched a free demo of SDXL Turbo on ClipDrop where you can try out this new model and generate images.
The model is also available for download for free under a non-commercial license allowing you to run it locally on your computer.
There are many web-based interfaces for Stable Diffusion that you can install locally on your Windows or Mac devices. Some notable names include Automatic1111, ComfyUI, InvokeAI, and DiffusionBee.
You can install the SDXL Turbo model on one of these interfaces and experience it first-hand.
Since SDXL Turbo is a single-step image generation model, it works even on devices with lower-end GPU specifications.
However, the output quality isn’t as detailed or high quality as the previous Stable Diffusion XL (SDXL) or Stable Diffusion 1.5 models.
The trade-off comes with a significant improvement in image generation times which can be quite useful in many cases.
For instance, real-time image generation with SDXL Turbo can be very useful during brainstorming and exploring ideas for an image. Once a good image is generated using SDXL Turbo, you can then switch to other models and generate the same image using the same prompts at a higher quality.
This can prove very useful to artists, designers, and anyone in the creative world who wants to quickly whip up a few AI images without going through a tedious process of spending hours generating the right image.
Stability AI Keeps Pushing The Landscape Of Generative AI
SDXL Turbo isn’t the only model released by Stability AI this past week. The company also released a bunch of new models including one for AI video generation.
Their new model named Stable Video Diffusion pushes the envelope of AI video generation forward as it allows users to generate photorealistic AI videos.
The company has released its image-to-video model and has opened a waitlist for its text-to-video generation model as well.
The models seem to surpass other models developed by competitors such as Pika Labs and Runway ML.
However, Stability AI has issued a statement saying that this model is only for research purposes and not real-world or commercial applications.
Apart from this, the company now has models for generative AI audio, image, and video covering the entire scope of the industry.
How Long Can Stability AI Keep This Up?
In the span of one year, Stability AI has gained over 10 million users and the company is valued at USD 1B. However, things aren’t going that great behind the scenes as many insider reports suggested that the company has burned through the USD 100M in funding it received recently.
Moreover, they have also been seeking to raise another round of funding at a USD 4B valuation. While giants such as Microsoft are pouring billions into their competitor Dall-E, Stability AI seems to be struggling to raise more money.
That’s why, the company recently introduced Stability Memberships which is a subscription that includes all the core Stable Diffusion models for commercial usage.
The company has set a revenue threshold similar to gaming engines like Unreal Engine where users will have to pay for the models after reaching a certain revenue.
This change comes largely so that the company can continue its research into Stable Diffusion and continue creating more models in the future.
It’s impressive how Stability AI is making continuous efforts to improve Stable Diffusion and change the landscape of generative AI. Despite the problems happening within the company, they seem to stay ahead of the industry when it comes to generative AI.
Time will tell how long they can keep this up and what else the company has in store for us in terms of Stable Diffusion and its expansive portfolio of models.
(ADVERTORIAL DISCLAIMER: The above press release has been provided by ATK. ANI will not be responsible in any way for the content of the same)
