Stable Diffusion Versus Disco Diffusion

There are numerous programs and solutions out there that aim to leverage Artificial Intelligence (AI) as well as Machine Learning (ML) as a means of enabling users the ability to essentially turn their words into artworks–all of these AI programs share many similarities but are fundamentally different in what they generate in an image in terms of quality.

Today, we'll go over Stable Diffusion versus Disco Diffusion to see which one stands out.

Disco Diffusion Explained

Disco Diffusion is essentially software that is written in the Python programming language, and its main goal is to run in Google Colab as a free tool that can be used to create art generated by AI from simple textual descriptions. In order to use it, users are required to use the Chrome Browser, have a Google Account, and activate their Google Drive. 

Google Colab is a free cloud service that is hosted by Google and aimed at encouraging AI as well as ML research, where the barrier to learning and success is the cost of hardware, normally. Google Colab provides a Jupyter Notebook environment that has no setup requirements, runs on the cloud directly, and is completely free.

Stable Diffusion Explained

Stable Diffusion is a model that is used by many online for the process of creating virtual artworks by leveraging the power of AI. It is fully open-source and employs a frozen CLIP ViT-L/14 text encoder, which is what enables its technology to utilise text prompts for creations.

It uses an image generation process during which it employs “diffusion” at runtime. Here, it starts an image off with noise and works its way towards removing it, so it can generate an image that is as close as possible to the original description.

Differences Between Disco Diffusion and Stable Diffusion

Each of these technologies has its own pros and cons, but let’s explore why Stable Diffusion typically works better than Disco Diffusion.

The general coherency and structure within the creations Stable Diffusion generates are much better, resulting in the ability to create more complex images. Additionally, Stable Diffusion is much quicker when compared to Disco Diffusion, where if a user has a Graphics Processing Unit (GPU), for example, with eight gigabytes of VRAM, they can run it directly on their PC for free and will only need a few seconds to generate an image. However, the Disco Diffusion model still makes amazing images but specialises in abstract imagery, which uses deep and vibrant colours alongside grainy imagery to create unique pieces. 

In this spectrum, Stable Diffusion does best in terms of creating a realistic image–in the form of an oil painting, for example; this means that it can specialise in the creation of artworks that are not photo-realistic but feature their own distinctive styles.

While Disco Diffusion can potentially be a lot cheaper to get into, Stable Diffusion is also open-source. However, it runs on the user's hardware directly, so it has hardware requirements associated with it. 

The Future With Either Option

No matter which one of these options a user ends up picking, it is clear that both of them play their own role in the text-to-image creation sphere–both options have their place in the digital artwork market, and it is likely that we will see a much higher level of competition going forward. 

There are other solutions specifically developed to solve issues and achieve better results in terms of AI-generated images, but they all do things a bit differently and, as such, should be utilised and experimented with to get the best possible outcome and turn an idea into a realistic artwork or image.