Fractal AI: Kalaido.ai paints a pretty picture from local language text prompts

“Kalaido, which was trained on a public dataset of 70 million images, has shown 40% higher efficiency in producing rich detail images than our global competitors during internal trials,” Fractal CEO and cofounder Srikanth Velamakanni told ET.

Fractal joins the AI diffusion model race with heavyweights such as OpenAI’s Dall-E, Stability AI’s Stable Diffusion, Midjourney and Runway among others.

Kalaido, however, is the first Indic language image model which can generate images and also enhance user prompts automatically to suggest improved descriptions, cause 2X reduction in time spent on iteration, and maximise creative output.

Also read | Ola founder Bhavish Aggarwal’s Krutrim AI turns unicorn with $50 million funding from Matrix, others

The model, which is currently available in beta version, is set to be launched by the end of this month for free-to-use, Velamakanni said.

Discover the stories of your interest

Its software code and weights will not be open-sourced, he said.“Kalaido was built on an ‘efficient training’ approach and is, therefore, capable of producing rich detail images in fewer steps, saving time, GPU costs, and reducing carbon footprint,” the CEO said, adding that the tool will be resourceful in industries such as advertising, graphic designing, social media marketing, and edtech.

For instance, for a global soup brand, Kalaido has led to a 79% reduction in the cost of generating an image, which would otherwise cost around €5,000, Velamakanni said.

“For a multinational consumer brand, we used neuroscience-based prompt strategy to understand decision-making in the brain and combined that with brand guidelines,” he said. “And, finally, we generated a 100-word prompt for text-to-image model which is then fed to Kalaido for generating images.”

The model can be fine-tuned to abide by brand guidelines, he added.

Kalaido has also helped an edtech company to generate 9-hour-long course content with just two hours of actual shooting. The rest was generated through images.

About practicing safeguards in publishing AI image generation tools, Velamakanni said the company is exercising extreme caution like other global leaders.

“We understand that a free-to-use tool like Kalaido can be misused for certain malpractices, especially deepfakes. For this reason, we are watermarking all images with kalaido.ai currently,” he said. “We also have AI capabilities which decline to process text prompts which have the potential to cause harm and we will improve these capabilities with time.”

Mumbai-based AI analytics startup Fractal, backed by TPG Capital, reached a valuation of $2 billion in 2022 when the latter infused $360 million into the company.

Among other generative AI capabilities, Fractal has launched FlyFish powering conversational commerce and the Marshal bot, which can assist business leaders and C-suite executives in critical thinking. It is an AI avatar of Marshall Goldsmith – top ten business thinkers in the world and an executive coach.

“Kalaido, which was trained on a public dataset of 70 million images, has shown 40% higher efficiency in producing rich detail images than our global competitors during internal trials,” Fractal CEO and cofounder Srikanth Velamakanni told ET.

Fractal joins the AI diffusion model race with heavyweights such as OpenAI’s Dall-E, Stability AI’s Stable Diffusion, Midjourney and Runway among others.

Kalaido, however, is the first Indic language image model which can generate images and also enhance user prompts automatically to suggest improved descriptions, cause 2X reduction in time spent on iteration, and maximise creative output.

Also read | Ola founder Bhavish Aggarwal’s Krutrim AI turns unicorn with $50 million funding from Matrix, others

The model, which is currently available in beta version, is set to be launched by the end of this month for free-to-use, Velamakanni said.

Discover the stories of your interest

Its software code and weights will not be open-sourced, he said.“Kalaido was built on an ‘efficient training’ approach and is, therefore, capable of producing rich detail images in fewer steps, saving time, GPU costs, and reducing carbon footprint,” the CEO said, adding that the tool will be resourceful in industries such as advertising, graphic designing, social media marketing, and edtech.

For instance, for a global soup brand, Kalaido has led to a 79% reduction in the cost of generating an image, which would otherwise cost around €5,000, Velamakanni said.

“For a multinational consumer brand, we used neuroscience-based prompt strategy to understand decision-making in the brain and combined that with brand guidelines,” he said. “And, finally, we generated a 100-word prompt for text-to-image model which is then fed to Kalaido for generating images.”

The model can be fine-tuned to abide by brand guidelines, he added.

Kalaido has also helped an edtech company to generate 9-hour-long course content with just two hours of actual shooting. The rest was generated through images.

About practicing safeguards in publishing AI image generation tools, Velamakanni said the company is exercising extreme caution like other global leaders.

“We understand that a free-to-use tool like Kalaido can be misused for certain malpractices, especially deepfakes. For this reason, we are watermarking all images with kalaido.ai currently,” he said. “We also have AI capabilities which decline to process text prompts which have the potential to cause harm and we will improve these capabilities with time.”

Mumbai-based AI analytics startup Fractal, backed by TPG Capital, reached a valuation of $2 billion in 2022 when the latter infused $360 million into the company.

Among other generative AI capabilities, Fractal has launched FlyFish powering conversational commerce and the Marshal bot, which can assist business leaders and C-suite executives in critical thinking. It is an AI avatar of Marshall Goldsmith – top ten business thinkers in the world and an executive coach.