Stable Diffusion
Day 7 / 366
The image you see above is AI-generated, but if you saw it anywhere on the internet you might easily mistake it for a screencap of a movie or TV Show. And I was able to create it in 10 seconds using Stable Diffusion. It’s crazy how convincing AI models are becoming in producing images, and I want to learn how to use them properly.
There are several popular models for generating images. We have DALL-E by OpenAI (The guys that made ChatGPT). That’s the model that is used in the image generator I created a few days back -
Then there is MidJourney, one of the most popular models among content creators. The only downside I found is that you can only use it through Discord.
But both of these models are closed models, i.e. A company owns their parameters and you have to send your prompt to them, which they will use to run the model and return the generated image to you.
This means 2 things -
- Generating images using these models is not free.
- The company would have full control over what sort of images can and cannot be created. The only control you have is the prompt.
This is where Stable Diffusion comes in. This is a model generated by Stability AI that you can download and run locally on your machine for free!
Not only that, you can fine-tune the model for your use case. For instance, you can train it on your selfies and use it to create realistic photos of yourself in any pose, in any location that you like!
Of course, you would require a pretty good graphic card with loads of VRAM to run it. But if you just want to try it out, you can use services like https://app.rundiffusion.com/ to rent some GPUs to run your SD models. That’s what I did to create that astronaut image.
Stable Diffusion has been around for more than a year, but I always avoided it thinking it's too tough to understand. LLMs on the other hand were easy. But now I have decided to get into SD and images as well, as there are lots of potential products that can come out of it.