Fractal (www.fractal.ai), a leading provider of artificial intelligence and advanced analytic solutions to entities including Fortune 500® companies has announced the launch of Kalaido.ai, India’s first text to image diffusion model.
Trained on a public dataset of 70 million images, Kalaido.ai takes text inputs in 18 different Indian languages to generate images. During internal trials, the tool has shown 40% higher efficiency in producing rich detail images that some of other existing solutions. It helps generate images of any desired style treatment (cinematic, illustrative, etc.) of any desired aspect ratio (1:1, 16:9, etc.) with just one click. Further, only open-source images have been used to train the model avoiding any chance of any copyright infringement and it is tuned to not allow Not Safe For Work (NSFW) outputs.
Using proprietary diffusion pipeline involving both “training free” approaches and minimal training, the tool helps to enable richer details in the image, higher fidelity and photorealism – delivering desired images in fewer steps, saving time, GPU costs, and reducing carbon footprint.
For a global soup brand, Kalaido has led to 79% reduction in the cost of generating an image which would otherwise cost around Euro 5000. The team used neuroscience-based prompt strategy to understand decision making in the brain and combined that with brand guidelines to generate a 100 word text prompt which was then fed to Kalaido.ai for image generation.