Australian AI StartupsLeonardo.ai is called a "substitute version" by users because it integrates multiple AI image production functions and allows creators to use it for free every day.Midjourney” and “Simplified version of Stable Diffusion”.
It is reported that one year after the company's product was launched, the number of users exceeded 7 million, the platform generated more than 700 million images, and it was favored by investors.
On December 7, 2023, Leonardo.ai announced that it had received US$31 million in seed round financing. Investors included Blackbird, Side Stage Ventures, Smash Capital, TIRTA Ventures, Gaorong Capital and Samsung Next. As of now, it has raised a total of two rounds of financing with a total amount of US$38.8 million.
In 2022, JJ Fiasson, Jachin Bhasme, Sami Ede, Ethan Smith, and Chris Gillis co-founded Leonardo.ai. Fiasson has served as an executive in two companies, working in hosting services, loan mortgage platforms, and Web3 game companies. The former company was acquired by Spirit Telecom after 12 years of operation. In 2021, he founded the game company Raini Studios, which failed after two years of exploration. In the process, he accumulated experience in generative AI and open source tools, so he turned to AI tool platforms.
It is understood that Leonardo.ai currently targets users in the gaming, advertising, fashion and architecture industries, and users can create and save multiple assets online. Fiasson said: "We will also do more to make AI creation more powerful in terms of controllability and fidelity, and it can be used out of the box."
“Simplified version of Stable Diffusion”
Currently, the use of mainstream AI image generation products requires two thresholds: paid subscription or technical equipment. For example, Midjourney charges $10-60 per month for limited generation, and Stable Diffusion requires a professional-level computer graphics card configuration.
Some "alternative" AI tools meet the lightweight needs of some users who want free daily trials without the need for software installation, including Leonardo.ai.
In terms of price, Leonardo.ai is more favorable than Midjourney, with prices ranging from US$10 to US$48. It also provides tourists with the opportunity to try it for free. 150 tokens are given for daily login, which can meet the needs of about 5-8 AI raw pictures.
In terms of functions, Leonardo.ai is based on the Stable Diffusion open source model, and the integrated plug-in supports AI image generation, image modification, image expansion and 3D material generation. Due to its simple operation and friendly interface, users have rated it as a "simple version of Stable Diffusion" with a high cost-effectiveness.
The Leonardo.ai Finetuned Models column is consistent with the tone of the CivitAI community. There are 36 types of fine-tuned models provided by community creators and platform officials, with styles including fantasy, surreal, pixel style, cartoon, etc.
After selecting the fine-tuning model, the user enters the editing interface, selects the running style, and generates the image using the prompt words. In addition, Leonardo.ai launched a new function "Add elements" supported by LoRA, allowing users to select multiple elements and adjust the weights to obtain a better style effect. Users can generate 1-8 images at a time, and adjust the image size, image quality, and background seamless connection (Tiling) by themselves.
Template images alone are not enough for Leonardo.ai to attract millions of users in a short period of time. It combines the practical advantages of other AI products in terms of prompt words, dynamic adjustment, model training, etc.
Users can randomly generate prompt words like rolling dice. In the style model, users can let the platform generate prompt words at will and modify and fine-tune them on this basis, or they can enter a basic description and let the system generate 2-8 prompt words for selection. After obtaining the image, users can remove the content in the image by reverse inscription. This function is similar to the prompt word plug-in of Stable Diffusion WebUI.
In addition to text prompts, Leonardo.ai also uses the image prompt function of Stable Diffusion, combined with controlNET analysis, and uses "pose to image", "edge to image" and "depth to image" to make the character in the picture consistent with the pose, shape or depth of the example picture.
Leonardo.ai has a high degree of freedom in dynamic image adjustment. After obtaining the generated image, users can continue to make fine adjustments, such as image enlargement and enhancement, and background removal. There are two other real-time modification functions. "Real-time generation" allows users to generate new images while entering prompt words. "Realtime Canvas" allows users to generate high-quality renderings while changing sketches. During the trial, Javi, the founder of Magnific AI, demonstrated how Leonardo.ai can turn a sketch into a finished image in real time.
In addition, its Motion feature allows users to upload static images and generate cool dynamic videos, which requires payment.
Its other canvas editor is similar to Runway's canvas content completion function, allowing users to expand the image within a range. In addition, the canvas can be erased with one click to remove characters and backgrounds, which is similar to the graffiti brush of Stable Diffusion.
In addition to the advantages of other products, Leonardo.ai also has a unique advantage - allowing users to train and fine-tune models to generate customized style images. According to Discord, users upload 8-15 directional style pictures for AI to learn. The self-training model will generate images based on the user's key prompt words, and the training time is about 30 minutes. Fiasson said: "This model is a private model unless you want to make it public on the platform. You can continue to train with the semantic understanding of the large model to fit the style features you want."
Currently, Leonardo.ai has opened up five functions: image generation, real-time canvas, screen editing, motion lens and real-time editing. There is also a 3D texture generation function that is still in the Alpha stage. Each set of pictures generated costs 4-20 tokens depending on the usage model and function.
Users can save, edit and build multiple materials such as style pictures, game assets and 3D modeling in the same style and feature set on Leonardo.ai. Moreover, these materials can be reused on the platform, attracting many creators of game design or video production.
Over 7 million users in one year
Initially, Leonardo.ai focused on content creation for gaming assets, but later expanded to various verticals because Fiasson first built a Web3 gaming company when he was exploring generative AI.
Fiasson graduated with a bachelor's and master's degree from the University of Sydney, Australia in 2013. After graduating from university, he founded a managed service IT company in Sydney, Cloud Business Technology, which was later acquired by the listed company Spirit Telecom. After graduating with a master's degree, he co-founded Loan Base, developing smart technology to change the way homeowners apply for mortgages. It is said that the company has helped solve nearly A$1 billion in loans and was acquired by Exit Solutions Pty Ltd in 2021.
In June 2015, Google released Deep Dream, which is based on the artificial neural network algorithm. It was originally derived from engineers' research on visual classification technology. The images it generates are often composed of animal faces, dizzying colors and intricate buildings. Later, some technology-savvy artists used Deep Dream to create outstanding works and use them for exhibitions, which showed that artificial intelligence networks could create works of art with unique value - this attracted Fiasson to study AI raw images.
“Whether for personal or professional use, the user community exploring creativity is rapidly expanding, and the emergence of artificial intelligence allows you to not have to learn a technology, which is what makes the artificial intelligence revolution really interesting.”
Fiasson recalled in the podcast that he had no artistic ability at the time and could only take simple photos. However, the works he created with Deep Dream were psychedelic and interesting, and they sold for a good price. He saw the commercial potential of AI raw images. After observation, he first set his sights on the Web3 game field, which lacked a sense of experience. In 2021, he founded Raini Studios to explore it in depth and served as a game operator.
After Midjourney came out, Fiasson led the team to try generative AI. During the operation, he needed to get involved in artistic content creation and encountered difficulties in ensuring the delivery time and quality of game works.
“We see the potential, but we also see the limitations.”
The gaming industry requires a unified style and main visuals, but the images generated by text prompts at the time were unstable and could not produce satisfactory works in batches. Moreover, new concept maps and creative visuals were needed for game iterations, which were beyond the control of text prompts. Fiasson's first exploration of AIGC was not successful.
In August 2022, Fiasson was researching Google's AI painting tool Disco Diffusion on the Internet. The article "A Guide for Latent Space Travelers" published by artist Ethan Smith on Latent Space at that time was called a classic tutorial on Disco Diffusion. In it, he explored how to use artificial intelligence to build an art form, and Sami Ede helped him branch out Disco Diffusion into MathRock Diffusion. Fiasson connected with them through the Internet and found that they were interested in exploring the use of generative AI to focus on game material creation, and they hit it off.
Fiasson called on former Loan Base co-founders Jachin Bhasme and Chris Gillis, former studio software engineer Peter Runham, and Sami Ede and Ethan Smith to found Leonardo.ai, aiming to shorten the "distance" between AI content creation and user creativity.
In the early stages, the Leonardo.ai team researched and established a complete 2D creation process that supports vector graphics output. Fiasson used a video to show that he completed the visual materials required for a Tetris game with fruit jelly elements in 10 minutes, and edited the game background using the canvas function in the test. This time, the color and texture of the AI-running image remained consistent.
In December 2022, their product was launched at the same time as ChatGPT, and a list of 20,000 early access users was released. In February 2023, several YouTube bloggers made recommendation videos, which received considerable views, especially videos with titles such as "Midjourney VS leonardo.ai" and "Midjourney Free Edition", which received over 100,000 views. This led to a snowball growth in Leonardo.ai's traffic.
"At that time, we had about 50,000 to 60,000 daily active users."
According to Writerbuddy's statistics as of August 2023, Leonardo.ai's average monthly traffic increased by 2.2 million and its net increase was 17.5 million in 9 months, ranking 7th in the top 50 most visited AI tools in the world. During the same period, Midjourney's traffic decreased by 8.7 million.
In one year, their users grew from thousands in the early days to over 7 million. According to similarweb, Leonardo.ai had 15.16 million website visits in December 2023.
As the platform's use cases increased, Fiasson discovered that "although our main focus is game content production, our AI tools may be applicable to a wider range of fields." He continued to integrate other AI product functions into the platform, adding a canvas editor as a drawing area, adding text prompts to generate 3D textures, etc.
"Obviously, we use a lot of open source technology to package applications on the platform, but we also want to build our own barriers." Fiasson believes that the products of existing companies are aimed at the mass market and lack user control and customization of images, so he integrated his own models with image models such as SDXL0.9, Stable Diffusion2.1, and Dream ShaperV7 in Leonardo.ai. It is said that Leonardo.ai distinguishes itself from software such as Midjourney, Adobe Firefly, and BlueWillow by giving users full creative control.
In addition, in terms of team building, Fiasson collected talents from all over the world. "Australia is not like Silicon Valley, there are not so many sophisticated startups, so our project has attracted the interest of most AI talents in Sydney." He also believes that remote work is very effective. It is reported that Leonardo.ai's development team is in Australia, the marketing team is in Central America, and co-founders and main engineers Ethan and Sami work remotely in Florida, USA and Hamburg, Germany respectively.
Fiasson is the CEO and product manager. He understands that remote work requires close online contact with designers and engineers to promote project development, but he emphasizes that Leonardo.ai does not adopt an old-school management model. It does not advocate overtime or over-manage employees, and uses this atmosphere to retain outstanding talents.
Over 700 million images generated
Since its launch, Leonardo.ai has generated over 700 million images within the platform and has worked with creators in industries such as entertainment, fashion design, advertising, architecture, and more. It claims to be currently being used by some of the top 3A gaming studios as well as Razer Axon.
“The uses for Leonardo.ai are endless,” said Fiasson. “We are trying to reduce the friction between ideation and creation, and inspire creativity.”
In April 2023, after Leonardo.ai experienced its first large-scale user growth, Fiasson began to pay attention to the user's onboarding process. He believes that teaching users how to interact withAI TechnologyOnly by interacting with each other can they discover more possibilities of the platform.
Javi and Emilio, founders of AI image enhancement company Magnific AI, are angel investors in the company. Javi said: "They have not only developed a very good user interface, but also gathered a huge community."
Discord is the core community for the company's business growth, with about 180,000 active creators. Leonardo.ai invites creators from many countries to jointly create tutorials, share usage materials and teaching links, and attract more users to join in.
It makes sense for Leonardo.ai to let users familiarize themselves with the platform through community teaching. The number of integrated functions increases every month, and it is still being tested and updated.
According to its community announcement, Realtime Canva, which Leonardo.ai tested in December 2023, was officially launched the following month, and the important function Motion was updated and launched. In January 2024, Realtime Gen was launched for instant image generation.