Text-to-image generators are a relatively new thing in the AI world. The idea of ‘feeding’ a program just a few words and it generating an image seems really exciting.
That is exactly what OpenAi’s DALL-E or Google’s Imagen can do. Users can input a short description of whatever they want to see generated, and the program shows an image.
As you can see, some results are incredibly accurate and detailed. However, it is important to remember the images above were previously chosen by humans, meaning there are probably a lot more images out there that are not as good.
Both DALL-E and Imagen are not accessible to the public and there is a reason for that. With such programs available universally anyone could create, for instance, fake news or an image with the intent to harass someone. Moreover, it is crucial to remember that most algorithms are trained on biased data. That means that the output can often be, for example, racist or sexist.
I believe that programs like DALL-E and Imagen will eventually become public. Hopefully, by then, we will figure out a way to make them less biased and less harmful.
Do you think that such programs will be available to the public in the nearest future? Do you think they should be?
Resources:
- https://www.theverge.com/2022/5/24/23139297/google-imagen-text-to-image-ai-system-examples-paper
- https://openai.com/blog/dall-e/
- https://imagen.research.google/
Projects Imagen and Dall-E are really cool to follow and see the outputs of these models, which are just astonishing. However, I wonder if artists would use that for “generating” artworks, who should have copyright to it? The artist who just created the prompt for the model, OpenAI, the whole internet, or maybe Dall-E/Imagen? Also I wonder if thanks to tech like this, the work of artists in the future will be just thinking about most creative prompt for the model to paint sth original – aren’t those models threat for painting industry?
Really amazing project! I am curious how this exactly works and how the program creates the final picture. I think that in the future it can be used for example as a new way of creating more interesting presentations with dedicated icons and pictures.