Generative artificial intelligence startup Leonardo has launched a new “universal upscaller” tool that can enhance the quality and resolution of any photograph or drawing.
This is the latest in a growing line-up of new tools built on the Leonardo platform including generative video, texture generation for 3D models and real-time image creation.
The Universal Upscaler does what it says in the title, you upload a photograph or image — including those made by AI — and it scales it up and improves the overal quality.
I tried it on a 512x512 pixel image of a cat on the moon, created using Realtime Generation, a very fast but low quality AI image generation tool. The results were impressive.
How does Universal Upscaler work?
When signed in to Leonardo.ai go to the Universal Upscaler link on the left-hand menu. Select an image to upload and click upscale.
You can change the creativity and upscale level which changes how varied the final output is compared to the original. A higher level of creativity will change the image more, whereas setting it to zero will simply increase the quality without changing the picture.
You can upscale even further by running the upscaled image back through the Universal Upscaler — although the higher the resolution of the input image, the more it will cost to upscale and the more varied the output.
How well does it work?
To put Universal Upscaler to the test I uploaded a series of different images with a range of output quality and detail. This included AI generated images and photographs.
The first U.S. president to be photographed
John Quincy Adams was the first U.S. President to be photographed. The sixth president had his likeness captured in March 1843 aged 75 and while serving as a Represenative of Massachusett's 8th district in Congress. This was 14 years after his presidency. The image is held by the Met Museum and is public domain.
A cat on the Moon
One of my favorite things to have AI generate is a picture of a cat on the Moon as its output and interpretation can vary so widely. This image was made using Leonardo’s Realtime Generation that creates very low resolution pictures but does it rapidly. I fed it into Upscaler twice and got an impressive, very realistic output.
My dog Oscar
Oscar is an energetic five year old cockapoo that loves his toys and is surprisingly difficult to photograph. I have this snap of him cuddling a soft bunny but it was blurry. Upscaler did a surprisingly good job of picking out some key details.
AI generated battle scene
Finally I wanted to test it on a more complex image. For this I had DALL-E 3 generate a picture with two soldiers talking in the foreground and a complex, busy battle scene in the back. I then gave Universal Upscaler the image and the prompt “make the background scene clearer”. It did a fantastic job of not just upscaling the soldiers but highlighting the action.
More from Tom's Guide
- ChatGPT finally has competition — Google Bard with Gemini just matched it with a huge upgrade
- Google Gemini: Everything we know about the advanced AI model
- I test AI for a living — here’s why Google Gemini is a big deal
