Meta has introduced “Imagine with Meta AI,” a novel AI image generator based on its Emu image synthesis model, crafted from a training set of 1.1 billion publicly accessible images from Facebook and Instagram. This standalone website enables users to transform prompts into unique images and is an extension of the technology previously available within messaging and social networking platforms.
The extensive dataset for Emu’s training includes images from publicly visible Facebook and Instagram accounts, potentially involving user-uploaded content. This approach has raised discussions regarding privacy implications, reminiscent of the adage “If you’re not paying for it, you are the product.” Despite the vastness of Instagram’s daily uploads in 2016, the dataset for training the AI model constituted only a fraction of the platform’s entire photo library.
Meta asserts that it relies solely on publicly accessible photos for model training, indicating that setting one’s photos to private on Instagram or Facebook may prevent their inclusion in future AI model training, unless the policy changes.
Similar to other AI image generators like Stable Diffusion and DALL-E 3, Imagine with Meta AI leverages the AI model’s learned visual concepts from the training data to create new images based on textual prompts. Access to this platform requires a Meta account, importable from existing Facebook or Instagram profiles, generating four 1280×1280 pixel JPEG images per session, adorned with a small “Imagined with AI” watermark.
Meta’s expansion of access to Imagine with Meta AI beyond chats aims to appeal to creative enthusiasts, highlighting its foundation model, Emu. While the generator excels in creating photorealistic images, it presents limitations in handling certain prompts, media outputs, and text rendering compared to other AI synthesis models like Midjourney and DALL-E 3. However, it displays diversity in rendering people from various ethnic backgrounds.
The AI image generator demonstrates competency but falls within an average spectrum in the realm of AI image synthesis, showcasing potential while facing certain limitations in generating diverse outputs and handling specific prompts.
वि.सं.२०८० मंसिर २१ बिहीवार १४:४१ मा प्रकाशित