Generating Images With Text, how deep does it go?

We're getting better at identifying objects in photos by combining image processing and deep learning methods, but there's still room for improvement. Researchers from Google announced that they've created a model of text-to-image AI generators. It is an open-source model that converts text prompts into images in mere seconds. In their paper published on arXiv, Google describes how they developed the model using neural networks and word embedding techniques.

Deep neural network-based text to image generators are a hot topic in image generation these days. With the current level of photorealism and high-fidelity, they have the potential to bring a new generation of innovative products and services to market.

AI-generated image created from the prompt: “the relationship between mind and matter”

Generating images of scenes and objects with a textual description is the holy grail of AI research. A revoutionary text-to-image generator could change the way we get images online.

We've all had the experience of trying to find the right image for a blog post or article, but only finding something close enough.  So, in order to make it easier for people without access to Adobe Photoshop or similar tools to create ideal images for their content and social media posts, AI researchers are working on text-to-image generation. Even in cases where the model doesn't generate a perfect image, it will be useful in the generation of ideas for artists and designers that can be rendered further by humans.

These kinds of generative models are at the heart of technology that takes photos and, often times hilariously, adds visual details based on what its artificial intelligence algorithms identify.

Some tools you can try yourself to generate your own images using AI:

1. Dall-E 2

"DALL·E 2 is a new AI system that can create realistic images and art from a description in natural language."

"Teddy bears working on new AI research underwater with 1990s technology"

2. Midjourney AI

"Midjourney is an independent research lab exploring new mediums of thought and expanding the imaginative powers of the human species."

Prompts:

1. "handsome cute black dragon in snowy forest, dnd character, background focus, fantasy, magic, realistic textured skin, hawk feather, hawk wings, lizard legs, lizard pose, big eyes, clear clean, by lya kushinov, Avetetsuya Studios, Alexandra Fomina artstation, by Makoto Shinkai, digital 2D, matte painting — test — ar 9:14 — uplight — stop 90 — no dof"

2. "chrome wolf, glossy, metallic, neon, symmetrical, tribal patterns, realistic, unreal engine, octane, redshift, artstation, behance, — quality 2 — stylize 4500 — ar 9:16 — uplight"

3. "portrait of the cutest red fox ever, fluffy, photorealistic, soft lighting, unreal engine — ar 3:4 — uplight — stop 80"

3. Hotpot.ai

"Hotpot helps you create amazing graphics, pictures, and text. AI tools like AI Art Generator spark creativity and automate drudgery while easy-to-edit templates empower anyone to create device mockups, social media posts, marketing images, app icons, and other work graphics. Designs are free or $1 per graphic."

Hotpot AI

While still a long way from perfect, the researchers that produced these models have led to some significant strides towards the creation of high-resolution images. I am excited to see what the future will bring for these technologies as this is just a first step towards computer generated images using text prompts.