Leonardo.Ai Wants You to Be in Control of AI Image Generation [CNET]

The company behind the platform hopes natural language processing gives the tool an edge in a crowded sector.

It’s not hard to see why image generation startup Leonardo.Ai has the name it does.

“Da Vinci was both an artist and an inventor,” said CEO J.J. Fiasson. “I’ve always been fascinated by the intersection of technology and creativity.”

In the crowded field of AI startups, image generation is a popular focus. Leonardo.Ai’s competitors include OpenAI’s Dall-E 3 and Adobe’s Firefly, as well as Midjourney, Pixlr and Night Journey. And they all essentially have the same mission: to make the creation of images and videos as simple as typing (or saying) a sentence. 

When Leonardo.Ai was founded in October 2022, Fiasson said, the goal was to create a generative AI platform that gives users more control over images, thanks in part to a better understanding of natural language — that is, the way you and I speak.

Early features of the platform (also called Leonardo.Ai) included tools like style reference, content reference and character reference to help Leonardo.Ai understand exactly what you want, as well as an AI canvas for editing.

“These features collectively ensure that users can work with more natural language prompts and still achieve the detailed, high-quality images they desire, giving them full creative control over the entire process,” a spokesperson said.

The platform has 17 million registered users and saw 4 million to 5 million monthly active users in May.

Paying attention to prompts

According to the spokesperson, a forthcoming foundation model, Phoenix, provides control by “adhering to prompts with extreme accuracy.” (Prompt engineering, or the art of writing to communicate to a generative AI model precisely what you want, can be a delicate operation on any platform.)

Phoenix is the first model Leonardo.Ai has built in-house. It’s in previews, with a launch planned for later this month.

The model, like others, is able to generate images based on “long and detailed prompts” with a “heightened level of prompt adherence,” the spokesperson said.

Prior to Phoenix, Leonardo used open-source models based on AI-generated-art program Stable Diffusion. They were trained on publicly licensed images, Creative Commons images and synthetic data, or data that’s been generated by algorithms or through computer simulations to replace real-world data.

The startup also allows you to train your own model to better meet your creative needs. As of December, the startup said, 470,000 models like this had been created.

In late 2023, Leonardo.Ai released a generative AI tool called Realtime Canvas, which lets you paint in real time with AI augmentation.

Another relatively new feature includes Edit With AI, which lets you edit images using prompts.

The Realtime Gen feature generates images as you type.

“[It’s about] making some of the whole process very, very easy and straightforward,” Faisson said. “Making people feel like they’re in control … rather than making them feel like it’s spin the wheel and maybe you get something that you want, maybe you don’t.”

To start, you choose a preset model like Phoenix, enter a prompt and click Generate to see an initial set of images.

I asked Leonardo for a kitten, teddy bear and unicorn at the beach.

It suggested I modify the prompt with “instructive language to quickly iterate,” such as, “Add a red hat” or, “Make it vintage.”

You can also choose from presets like Anime, Graphic Design and Stock Photography.

It even offered to improve my query with an adjective-heavy prompt:

“A playful kitten, a cuddly teddy bear, and a majestic unicorn frolic together on the sandy shore beneath a bright sun. The kitten’s fur shimmers in the golden light, the teddy bear’s plush is pristine and inviting, and the unicorn’s horn gleams with magical energy. This enchanting scene is depicted in a stunning painting that captures the whimsical charm and innocence of these beloved creatures. The vibrant colors and intricate details make this image a true work of art that radiates joy and wonder.”

In her review of Leonardo.Ai, CNET’s Katelyn Chedraoui called the tool a solid option, on par with competitors like Firefly. She noted that it does indeed adhere to prompts better than platforms like Midjourney but lacks comprehensive editing tools and what she described as a “wow” factor.

Leonardo.Ai offers a free, limited version of the platform, along with premium plans that start at $10 per month. Features like character, style and content reference require a premium plan. So does training your own model.

The platform was initially targeted at the gaming industry, but that’s since broadened to include marketing and advertising; design; and entertainment, though Fiasson said the tool is also meant for the general public. Now Leonardo.Ai is developing a business-focused offering, which will allow enterprise users to train a model using their own datasets and then work together within the platform.

In December, Leonardo.Ai raised $31 million from investors like Blackbird, Side Stage Ventures, Smash Capital, Tirta Ventures, Gaorong Capital and Samsung Next.

“A lot of it is around, How do we actually make that more powerful, more controllable? How do we enable people to actually have a greater degree of control over the output?” Fiasson said. “And, How do we enable people to use tech to get from this starting concept to a very finished end point?”

On Tuesday, Leonardo.Ai announced it would be acquired by graphic design platform Canva.

This is one of a series of short profiles of AI startups, to help you get a handle on the landscape of artificial intelligence and the activity going on. For more on AI, see our AI Atlas hub, which includes product reviews, news, tips and explainers.