Dall-E’s new tool Outpainting can expand famous artworks by adding new elements
OpenAI has added a new feature called Outpainting in the Dall-E neural network, allowing users to expand the boundaries of an image using “visual elements in the same style,” the AI non-profit company announced in a blog post. Launched in January 2021, Dall-E can convert text-based instructions into images and artwork using GPT-3, a natural language processing (NLP) model.
NLP is a subset of AI, which allows the software to read, understand, and derive context from text and spoken words just like humans.
According to OpenAI, Outpainting can imagine what is outside the frame of an original image and add them to a larger version of it based on its interpretation of existing visual elements such as textures, shadows, and reflection. For this, it uses something called a natural language description.
The feature is available to all Dall-E users on desktop, the company added.
Dall-E already has a feature called Inpainting, which allows users to edit and make changes within a generated or uploaded image.
Currently available in beta, Dall-E was expanded to 1 million users in July. Its users include artists, researchers, and designers.
Dall-E’s creator Open AI is a non-profit AIM research company that was set up in 2015 by tech founders including Elon Musk of Tesla and Sam Altman, who is the current CEO, to make AI tools available for free to developers and researchers.
The recent advancements in NLP have allowed AI researchers to build neural networks that can recreate images or videos from text or a single image. For instance, last month, Alphabet company DeepMind unveiled a generative framework called Transframer that can generate a 30 seconds video from a single image. The framework uses a range of visual prediction tasks, including video modeling, novel view synthesis, and multi-task vision.