ImageFX is like OpenAI’s DALL-E. It can make whole images using what you tell it. But, is it good, and can you use it right away?

Google is a bit behind in artificial intelligence. While OpenAI launched the revolutionary DALL-E AI art image generator two years ago, Google has just released its own competitor.

ImageFX, supported by a major tech giant and vast amounts of data, raises the question of how well it performs.

In a nutshell, ImageFX is generating impressive images, standing up to the best. How does it operate? Can you use it now? And has it addressed key issues in the AI art world?

How to use Google ImageFX

As of now, Google ImageFX is accessible for use in several countries, including the US, Kenya, New Zealand, and Australia.

Attempting to access the site in countries like the UK will result in a prominent warning stating, ‘This tool isn’t available in your country yet.’

To use it from one of the currently supported countries, visit Google’s AI Test Kitchen, create an account, and once you’re set up, you can start experimenting with your prompts.

If you’re not in the mentioned countries, it’s still worthwhile to visit the website. Google allows you to sign up for notifications to alert you when the platform becomes available in your location.

How good is Google ImageFX?

There’s no denying that Google is late to the game. OpenAI’s DALL-E debuted in January 2021, followed by Midjourney a year later. The question now is whether Google’s delay has translated into a superior quality product.

The images released thus far indicate that ImageFX can create content at an impressive level. It’s detailed, understands context well, making ImageFX a predictably capable image generator.

Certainly, it’s a valid point. AI art has made remarkable progress, and Google’s major competitors are consistently producing work of comparable quality, having been in the game for a much longer time.

Currently, the significant advantage of ImageFX is its free accessibility (in select countries). Unlike Midjourney and DALL-E, which are often behind paywalls or restricted services, ImageFX offers a valuable opportunity that is worth taking advantage of before potential changes occur.

ImageFX has introduced a distinctive feature called ‘expressive chips.’ This feature enables users to promptly edit a prompt and explore different styles in their search. For instance, if you requested a portrait of a woman, you can swiftly switch it to an abstract, hand-drawn, or even an oil painting style.

How does it work?

In essence, Google ImageFX operates just like all other AI art generators. This involves a series of steps, beginning with acquiring a substantial database of images for training purposes.

Although Google hasn’t officially disclosed the origins of its training data, it probably stems from a blend of internal sources, collaborations with organizations and research institutions, and possibly includes web scraping and user-generated content.

After assembling the database, the model undergoes training on these images, establishing connections between words and visual concepts, likely utilizing a diffusion model.

These models initiate with random noise in an image and iteratively refine it, guided by information from both the data and accompanying textual descriptions. This repetitive process essentially teaches the model the associations between words, images, and context.

Through this training, the model gains the ability to comprehend which words are linked with an image. This understanding enables ImageFX, and any AI image generator, to interpret the given prompt effectively.

How is it linked to Google Bard?

Google Bard, considered by some as a significant rival to the AI chatbot ChatGPT, was kept under wraps for a while before its public launch in 2023.

In the realm of AI, ImageFX handles the visual, while Bard focuses on words and contextual understanding. The ultimate aim is to merge these two components to craft a comprehensive AI model, mirroring OpenAI’s goal of combining ChatGPT with DALL-E, its image generator.

Google Bard is currently undergoing a testing phase, but it’s anticipated to be fully accessible soon, supported by the newly introduced Google Gemini system. The potential implication is a platform where you could instruct the model to generate a complete board game, providing rules, lore, images, board, and content. Alternatively, you could task it with crafting a series of books, complete with illustrations.

Can ImageFX create inappropriate images?

AI art faces a challenge: people. Trained on artwork created by humans and then used by humans, there’s a risk of inappropriate elements reflecting certain aspects of the human mind.

Earlier AI art generators have exhibited issues like sexism, prejudice, and at times, highly graphic images. Addressing this concern is a shared goal among major tech companies, including Google with ImageFX.

Google emphasizes that all images produced with ImageFX bear a digital watermark embedded by SynthID, a tool developed by Google DeepMind. This serves as a distinctive mark on the generated content.

The SynthID watermarks added to the images by ImageFX are invisible to the human eye but are designed for detectability for identification purposes. Furthermore, each image comes with metadata, offering additional information to individuals encountering AI-generated content.

Google has affirmed efforts to enhance the safety of training data, resulting in reduced problematic outputs such as violent, offensive, or sexually explicit content. These improvements also extend to minimizing the potential to generate images of real individuals.