Google's AI photo editor lets you use words to describe what to edit - Android Police

🥇 Bonuses, Promotions, and the Best Online Casino Reviews you can trust: https://bit.ly/BigFunCasinoGame

Google's AI photo editor lets you use words to describe what to edit - Android Police

Text-based image editing is coming Source: Steven Winkelman Artificial intelligence and machine learning have been Google’s passion projects for several years now, and the I/O 2023 keynote address only made that more apparent. Image creation is one of the more intriguing applications for this technology, and Google’s efforts in this area materialized as Imagen, a text-based image generation tool much like Midjourney and DALL-E 2. Now, Google is sharing research showcasing Imagen Editor, where textual prompts and a little sketching can suffice to perform local edits on photos. Google’s Imagen utility is already adept at creating images from scratch, solely off of textual prompts. However, if you aren’t satisfied with the result, you’re usually forced to restructure your prompt, polish it, and give the image generator another go, simply because Imagen doesn’t yet allow editing specific elements of images you aren’t happy with. To address this, Google recently shared research for Imagen Editor and EditBench, utilities currently in beta, but capable of guiding edits with text prompts. Instead of creating fresh images using a prompt, Imagen Editor needs a photo that must be edited, a text prompt from the user defining the change, and a masked region defining where the edit needs to be applied. The result is edits limited to the region you defined, tailored to the prompt provided. Moreover, the results are photorealistic and natural. Masked region and Imagen Editor’s results for “a bouquet of red flowers,” “two trees,” “an Imagen Editor sign,” “a bush with green leaves,” and “a bush without leaves” Technically called inpainting, the process Google’s new tool uses is like an image restoration or something we can best describe as the confluence of Google AI and Adobe Photoshop’s Content Aware Fill. The researchers developed new encoders for Imagen Editor and also included an object detector module in the AI to compensate for incomplete or inaccurate masks. The research also includes a tool called EditBench to evaluate results of text-guided inpainting. Based on a 240-image dataset, the benchmark evaluated edits on both human-made and AI-generated images on parameters like the modified objects, their attributes like shape, size, number, and suitability for the scene. Google observed that object masking helps improve image-text alignment, making Imagen Editor better than alternatives like DALL-E 2 and StableDiffusion in all the categories EditBench tested. Unfortunately, Google has unspecified concerns related to the responsible use of AI, and that’s why it won’t be releasing Imagen Editor to the public. The company recently proposed a framework to safeguard AI development, and hopefully, a few hard limits can be established before giving people access to tools like Imagen Editor. On the bright side, EditBench is available in its entirety, for free, to help further AI research. Meanwhile, we remain hopeful the base model, Imagen, is soon integrated into Gboard.

#AINews, #ArtificialIntelligence, #FutureOfTech, #AIAdvancements, #TechNews, #AIRevolution, #AIInnovation, #AIInsights, #AITrends, #AIUpdates