A new command (“prompt”) arrives in Midjourney. It describes an image. Four proposals are made for each photo sent. Then, it is possible to create variants on the snapshot.
There are changes in Midjourney since April 4, 2023. The laboratory behind this artificial intelligence specialized in the production of images has sent an update to its system which will be of interest to those who use it. Midjourney now manages to describe an image sent to her. At the very least, he tries to analyze it correctly.
Describe an image with Midjourney
For this, Midjourney has added a new command (a “prompt”) that can be used at any time. Just type ” /describe (without the quotes) in the input field on Discord. This will bring up an action button through which you can find your image on your computer — if it’s an image that’s on the web, you’ll need to download it first.
Once your image has been submitted, Midjourney will display four proposals to describe the visual. Why four? Because artificial intelligence may not fall just by analyzing the snapshot. So she offers four variants by replacing a few passages each time – whether on the subject, the style or the technique of the photo.
In the example of this red panda, imagined by DALL-E, a concurrent generative AI, the four descriptions offered were:
- a red panda with stars in the sky, flying in the sky, in the style of dark magenta and light blue, playful, dreamlike imagery;
- the red panda flying in space with stars, in the style of gabriel bá, cryptopunk, brushwork mastery, wlop, oshare kei, ferrania p30, anna dittmann;
- red panda flying in the starry sky, in the style of gabriel bá, cryptopunk, sakimichan, dark magenta and light blue, robert munsch, brushwork mastery, ferrania p30;
- red panda flying through the night sky in the middle of a galaxy, in the style of brushwork mastery, comic art, caninecore, wimmelbilde.
Create image variants on Midjourney with suggested prompts
You then have the option of clicking on the description that seems most relevant to you. Midjourney will display a pop-up window in which it will offer you, if you wish, to modify the prompt or add instructions. For all practical purposes, the system prompts you to ensure that you do not transmit passwords or personal data.
Once the command is sent, Midjourney tries to recreate images (four, again) corresponding to the parameters you sent to it.
The last steps are the ones you already know if you’re used to Midjourney: you can generate four other visuals (if you don’t like any in the list). Midjourney also offers you to select one and to decline it in four variants. Finally, if one catches your eye, you also have the option of producing it in high quality.
More or less happy descriptions
We experimented three times with the command to see how Midjourney fared. The photos used here were in the public domain for the first and free of rights for the second. It shows Charlie Chaplin in 1914, in the film Charlie is happy with himself. The other features a SpaceX Falcon 9 rocket on its launch pad.
Here are the descriptions suggested by Midjourney and the illustrations generated according to our choices, without retouching them.
In this description, Midjourney had some difficulty interpreting the image, sometimes evoking a horse, sometimes a lion. He also said that Charlie Chaplin was here focused on children – which is incorrect. He nevertheless saw Charlie Chaplin and the crowd. We chose the first photo, but Charlie Chaplin disappeared in the four visuals offered.
In this other example, Midjourney fared better. He mentioned SpaceX and a rocket in his prompts (even if he spoke curiously of a Falcon…4, whereas it’s the 9). The colors have been well identified and the twilight atmosphere too. The last photo is the one that diverges the most, with a second rocket on the screen. Launchers have very varied profiles.
It is important to note that the images used in this demonstration are no longer or not covered by binding rules on copyright. If you use Midjourney, make sure you have the right to use these images in an AI – artists, indeed, oppose it.
If you liked this article, you will like the following ones: do not miss them by subscribing to Numerama on Google News.