Stable Diffusion Image Generator

Stable Diffusion is a state-of-the-art text-to-image art generation algorithm that uses a process called "diffusion" to generate images. "Diffusion" works by training an artificial neural network to reverse a process of adding "noise" (random pixels) to an image. Once trained, the neural network can take an image made up of random pixels and turn it into an image that matches your text prompt.

The makers of this model chose to make stable diffusion public only recently, and its release is a major landmark in the AI art space.

Stable Diffusion was created by researchers employed or sponsored by Stability AI and the CompVis team at Ludwig Maximilian University of Munich. The model training was funded primarily by Stability AI and sponsored in part by NightCafe Studio.

If you'd like to learn more about the model, the technical details are available on CompVis's stable diffusion GitHub profile.

In terms of image outputs, Stable Diffusion and DALL-E 2 are quite similar. DALL-E 2 is often better at complex prompts, while Stable Diffusion images are often more aesthetically pleasing. With just 890M parameters, the Stable Diffusion model is much smaller than DALL-E 2, but it still manages to give DALL-E 2 a run for its money, even outperforming DALL-E 2 for some types of prompts. Unlike DALL-E 2, the Stable Diffusion code and trained model is Open Source and available on GitHub for use by anyone.

In short, the two image generation models are comparable, with minor differences in how closely their outputs match expected outcomes.

It's free to generate up to 28 artworks per day using any of our image generation tools. To create more images, you can subscribe, purchase a credit pack, or earn credits by participating in the community. Subscriptions are not required to generate images.

NightCafe Creator is a web-based image generation app. You won't find it on any app store, but you can install it and run stable diffusion from the home screen of your iPhone, Android phone, or tablet. Once the image generation app is installed on your device, you can begin to create beautiful images from text.

Yes! As long as you own (or have permission to use) any images that you used in the creation process, we transfer any Copyright assignment to you - the creator. Please check the Copyright laws in your own country to confirm. Copyright and license laws for AI-generated art are still evolving, and some jurisdictions will not grant Copyright for AI-generated artworks.

As a latent diffusion model, the stable diffusion code creates images by removing noise through a series of steps until it arrives at the desired image. The technical details can't be glossed over, so if you want a more in-depth understanding of the process, you'll need to start by learning how convolutional networks, variational autoencoders, and text encoders work together in this type of machine learning model.

In slightly simpler terms, stable diffusion was first trained with a database of text and image pairs provided by Laion. Using this database the model was gradually “taught” how to generate images from a seed by starting with a rough result and gradually improving the resolution it until it satisfies certain conditions.