This year 2022 will go down in history for the advances that have occurred in the field of artificial intelligence. AI has made a qualitative leap in almost every area. However, neural networks have left us speechless for their ability to generate images from text. Do you want to enter this world? Today we’re going to talk about the options you have for doing this and what we consider to be the best tool to start generating art from text.
What imaging AIs do we currently have?
It’s not just a few tech companies that have invested all sorts of resources to achieve intelligence capable of creating art. Doing a quick review, here are the most relevant:
- Dall-E: this is the OpenAI proposal. This company has released many models over the past couple of years. The first variant of Dall-E can be used for free. The second, infinitely more powerful, can be used from its official website for a small payment for each generation.
- Mid-term: it is the second that has arrived. It is a really powerful AI, being its strong point the creation of artistic images. Some that we have published in El Output in the last two months are illustrated with images generated by this AI. Its only downside is that it has to be paid for – and it’s not exactly cheap.
- Stable Streaming: This is an open source model. Currently, it is the most accessible AI that exists, since it can be used completely free of charge.
Stable streaming: the best AI you can use to generate images
To be honest, I personally prefer Midjourney for its comfort and artistic results. However, Stable Diffusion managed to reach another level in a few months. So much so that it already far exceeds payment alternatives.
Being an open model, Stable Diffusion is not limited to a single development team. Instead, any programmer on this planet with the necessary knowledge can do their part. Despite the fact that this software was released very recently, there are already many applications that allow you to take full advantage of it.
The main features of Stable Diffusion are:
- text-to-image: this is the basic mode of use. You write a prompt and the AI will generate an image according to the description.
- image-to-image: Generates an image based on another we provide. The possibilities of this feature are limitless, being able to create an ultra-realistic photograph of something as simple as a child’s drawing.
- Local execution: only a computer with an Nvidia card supporting Cuda is necessary. It is recommended to use a GPU with a minimum of 4 GB of VRAM, although there are modified models that allow it to be used with less powerful hardware.
- Uncensored: By using the model locally —or changing some settings in Colab—, this software can be used without limitation.
- Custom Training: Although not an easy process, there is parallel software, DreamBooth, that allows us to train a custom stable broadcast model with the images we want. Thanks to this, you can have a model capable of generating images with your face, which none of the other AIs that compete with this technology can do so far.
- Third-party applications: new applications using this artificial intelligence are launched every week. There are utilities for using Stable Diffusion on Windows with different types of GUIs, as well as plugins for Photoshop.