This Tuesday a group of three images that represent a very improbable scene in current Spanish politics have gone viral. This is Pablo Iglesias and Yolanda Diaz together in various situations, all smiles between themafter the official break between the two that materialized this weekend with the launch of the political formation Sumar in which he does not participate Can. The images are obviously not real but a work carried out with artificial intelligence tools by the United Unknown collective for El Mundo.
The great challenge of artificial intelligence: “The quality of the truth is going to decrease very quickly and that is enormous damage for democracy” https://t.co/VX5czE8yeK
— THE WORLD (@elmundoes) April 4, 2023
Although fakes and doctored images are nothing new, even less in the field of disinformation and political communication, what the new text-to-image artificial intelligence tools are allowing is make available to anyone software that allows them to be created in a simple way and with a quality that, in some cases, can pass them off as real. This work by United Unknown, a group of political satire and humor video and image creators founded in 2010 and have collaborated with a wide variety of media since then, is a case in point.
However, images with this quality are not as easy to achieve as telling an AI from text to image “Pablo Iglesias and Yolanda Díaz walking down the street together and smiling”. For results that look almost photorealistic much more is needed.
In this case, as explained by United Unknow to El Mundo, they received the order last Friday and They spent the whole weekend to get the images that the media has published today. They did not use a single AI but 3: DALL-E, Stable Diffusion and Midjourney.
“Soon it will be impossible to distinguish what is true from what is false”
With the wonderful collaboration of @unitedunknown https://t.co/ch3fpUCTFi pic.twitter.com/ROjoj1Y0vp
— Rodrigo Terrasa (@rterrasa) April 3, 2023
The first is the work of OpenAI, the same company behind ChatGPT, which Microsoft has integrated with the new AI-powered Bing. Stable Diffusion is the least known of the three, although it has the advantage that it can be installed on a computer and run locally.
midjourney It is the most popular among the followers of this type of tool and with which work as viral as the false arrest of Donald Trump on the streets of New York has been carried out. In fact, Midjourney Inc. has decided to modify its business model and suspend the free modality of use of Midjourney to limit the creation of images that can be used as disinformation.
According to United Unknown, this work had the difficulty of having to faithfully recreate two known personalities instead of one and it was necessary generate between 100 and 200 images for each of the three final images. This was followed by significant post-editing work to process, readjust, retouch and obtain the final version.
It’s not that easy, but in 2023 the creation of these types of images is much more accessible than just two years ago, and the results of these tools are much better than what was possible just one year ago, when the first version of Midjourney was released.
It is true that “photographs” can be faulted, but seeing the path traveled by the AI in such a short time, it is difficult to predict what we will be seeing in another year. What is clear is that the Internet is going to be filled with images generated by artificial intelligence and that next on the list is the video.
Generate videos with nothing but words. If you can say it, now you can see it.
Introducing, Text to Video. With Gen-2.
Learn more at https://t.co/PsJh664G0Q pic.twitter.com/6qEgcZ9QV4
—Runway (@runwayml) March 20, 2023
Runwayfor example, is another AI that allows you to generate video from other images and whose new version, Gen2, it will also do it from text descriptions. Its launch is near and it will most likely continue to evolve as fast as text to image. What will we be seeing in one or two years on our screens? Probably, something that never existed.
Discussion about this post