Google is lagging behind in artificial intelligence. While OpenAI’s innovative Dall-E AI art image generator was released two years ago, Google only recently released its competing product.
The software, known as ImageFX, is backed by one of the largest technology companies and a substantial amount of data. So how is this data accumulated?
In brief, ImageFX has produced some impressive images that rival the best. But how does it work? Can it be accessed now? And have major problems in the AI art world been solved?
How to use Google ImageFX
Google ImageFX is currently available in countries like the United States, Kenya, New Zealand, and Australia.
If you attempt to access the site in a country like the UK, you’ll see a warning stating, “This tool is not yet available in your country.”
To access it from any of the currently available countries, visit Google’s AI Test Kitchen. Then create an account. Once everything is set up, your new prompt will be ready for use.
Even if you’re not in one of the listed countries, the website is still worth visiting. Google allows you to sign up for notifications about when the platform becomes available in your area.
How good is Google ImageFX?
There’s no denying that Google is late to the game. OpenAI’s Dall-E was released in January 2021, and Midjourney was released a year later. So did Google’s delay pay off in terms of quality?
The images released so far demonstrate that ImageFX is capable of producing content at a very high level. Detailed and contextual, ImageFX is an unsurprisingly capable image generator.
But that’s expected. AI art has made significant progress over the years, and Google’s main competitors are producing similarly high-quality work and have been doing so for much longer.
The significant advantage of ImageFX at the moment is that it’s free (in select countries). Both Midjourney and Dall-E are mainly behind paywalls or restricted services, so it’s worth making the most of ImageFX before any changes.
ImageFX also includes a unique feature called the “Expressive Chip.” This allows users to quickly edit the prompt and try a different search. For example, if you request a portrait of a woman, you can quickly switch this to an abstract, hand-drawn, or even oil painting.
How does it work?
Basically, Google ImageFX works like any other AI art generator. This involves several steps, starting with obtaining an image database large enough for training.
Google has not disclosed the source of its training data, but it likely includes a combination of internal sources, collaborations, and possibly web scraping and user-generated content.
Once the database is built, a model is trained on these images to learn the relationships between the words and visual concepts in the images, possibly through a diffusion model.
These models start with random noise in the image and are refined based on information from both the data and the accompanying text description. By repeating this process, you essentially learn the relationships between words, images, and context.
This training helps ImageFX and other AI image generators understand the prompts asked because it understands what words are associated with the images.
How is it linked to Google Bard?
Google Bard is probably the biggest competitor to the AI chatbot ChatGPT. Google has been working on the chatbot for some time and was released publicly in 2023.
If ImageFX is photography, Bard is understanding words and context. The goal is to combine the two to create the ultimate AI model, similar to OpenAI’s combination of ChatGPT and Dall-E (OpenAI’s image generator).
Google Bard is currently in testing but will soon be fully operational with the recently announced Google Gemini system.
This could theoretically mean a platform that asks models to create a board game and returns both the rules and lore, as well as all images, boards, and content. Or you could write a series of books with illustrations to go along with it.
Does ImageFX produce bad images?
There’s a problem with AI art…people. When trained on artwork from a human population and then utilized again by humans, less appropriate parts of the human brain tend to enter.
Previous AI art generators displayed sexist, biased, and sometimes intensely graphic images. This is a problem that all major technology companies are trying to tackle, including Google with ImageFX.
“All images generated with ImageFX are marked with SynthID, a tool developed by Google DeepMind that adds digital watermarks directly to the content we generate.” Google says:.
“SynthID watermarks are imperceptible to the human eye but can be detected for identification. Additionally, all images contain metadata, so when you encounter an AI-generated image, You can get more information.”
In addition to this, Google announced that it has improved the safety of its training data, reducing problematic output such as violent, offensive, or sexually explicit content. This extends to a reduced ability to create images of real people.
read more:
Source: www.sciencefocus.com
